The Effects of Targeted Recruitment and Comprehensive Supports for Low-Income High Achievers
at Elite Universities: Evidence from Texas Flagships
Rodney J. Andrews, Scott A. Imberman and Michael F.
Lovenheim
The Effects of Targeted Recruitment and Comprehensive
Supports for Low-Income High Achievers at Elite
Universities: Evidence from Texas Flagships∗
Rodney J. Andrews
The University of Texas at Dallas and NBER†
Scott A. Imberman
Michigan State University and NBER
Michael F. Lovenheim
Cornell University and NBER
September 2015
Abstract
We study a set of interventions in Texas that were designed to overcome the multitude of hurdles faced
by low-income, high-ability students in the higher education system. The Longhorn Opportunity Scholars
(LOS) and Century Scholars (CS) programs were implemented in 1999 and 2000, respectively, and involved
recruiting at specified low-income high schools, providing additional financial aid, and enhancing academic
supports once enrolled in college if students attended University of Texas - Austin (LOS) or Texas A&M College Station (CS). These two schools are the flagships of Texas public universities and are widely regarded
as the top public universities in Texas. Using administrative data on all public college students in Texas we
find via difference-in-differences estimators that these programs, and in particular the LOS program, had
large, positive effects on high achieving students. Both interventions increased attendance at flagships while
LOS also increased attendance at other 4-year schools at the expense of 2-year schools. The latter finding
indicates substantial spillover effects of the recruitment portion of the program to students who are not able
to attend the flagships. Reduced-form intention-to-treat results show that, amongst students in the top 30%
of their high school class who attend any college, BA attainment increases by 3.5 percentage points from LOS
while earnings increase by 4.4%. Upper-bound treatment-on-the treated estimates that attribute all of these
gains to flagship enrollment are 58 pp and 110%, respectively. There is no statistically significant impact on
these measures from the CS program, though we cannot rule out sizeable positive effects (upper bound TOT
earnings estimates are an insignificant 43%). These results indicate that school-based targeted recruitment
can substantially increase enrollment of low-income students in higher quality colleges and, when combined
with adequate support services, improves educational and labor market outcomes.
KEYWORDS: Postsecondary Education, Higher Education, Low-Income Students
∗ We gratefully acknowledge that this research was made possible through data provided by the University of Texas at Dallas
Education Research Center. The conclusions of this research do not necessarily reflect the opinions or official position of the Texas
Education Agency, the Texas Higher Education Coordinating Board, or the State of Texas. We would also like to thank Sara
Muehlenbein and Mark Lu for excellent research assistance. Finally, we’d like to thank seminar participants at the Association for
Education Finance and Policy Annual Meeting, Dalhousie University, Institute for Research on Poverty Summer Research Workshop,
Michigan State University, Middle Tennessee State University, Society of Labor Economists Annual Meeting, Syracuse/Cornell
Summer Education Seminar, University of Michigan, University of Rochester, and Vanderbilt University for helpful comments.
† Corresponding author contact information: Rodney Andrews, Department of Economics, University of Texas at Dallas, 800
West Campbell Road, WT21, Richardson, TX 75080; email: [email protected].
1
Introduction
Changes in the US economy over the past several decades have led to historically high demand
for skilled labor (Autor 2014; Autor, Katz and Kearney 2008). In 1979, the gap in median yearly
earnings between households with at most a high school degree and households with a worker
who has a college degree was $30,298. By 2012, this gap had nearly doubled to $58,249 (Autor
2014). The increasing earnings premium associated with having a college degree underscores the
immense and growing importance of postsecondary education in driving labor market outcomes.
However, these high returns have been met with sluggish increases in postsecondary attainment,
particularly among students from low-income backgrounds (Lovenheim and Reynolds 2013;
Bailey and Dynarski 2011; Bound, Lovenheim and Turner 2010). For example, Bailey and
Dynarski (2011) show the college enrollment gap between those in the bottom and top income
quartiles grew from 39 percentage points to 51 percentage points between the early 1980s
and the turn of the 21st century. The college completion gap between these two groups also
grew dramatically during this period, from 31 percentage points to 45 percentage points. The
unequal investment in postsecondary education across the income distribution combined with
the large earnings premium associated with college graduation suggests the current higher
education system may contribute to, rather than mitigate, growing income inequality in the US.
Indeed, some evidence suggests that changes in the earnings premium associated with college
can explain between 60 and 70 percent of the rise in income inequality over the past several
decades (Goldin and Katz 2007). Thus, developing policies that can support the collegiate
attainment of students from low-income backgrounds is of primary policy importance.
Differences in collegiate investment between low-income and high-income students take two
forms. The first is that students from low-income families are less likely to attend college
at all (Bailey and Dynarski, 2011; Carneiro and Heckman, 2002). For example, tabulations
from the 1997 National Longitudinal Survey of Youth (NLSY97) show that while only 13%
of students from families with earnings over $125,000 do not attend college, 56% of students
from families with income below $25,000 do not attend college. As family income increases,
the likelihood of attending college increases steeply. The second type of investment gap, which
has received far less attention, is that low-income students tend to enroll in schools of lower
1
quality than their higher-income counterparts (Hoxby and Avery 2013; Lovenheim and Reynolds
2013). In the NLSY97, only 2% of low-income students attended a flagship public school,
while among the highest-income students 16% did.1 The likelihood of attending a private
school also increases with income, and the proportion of students enrolling in a two-year school
declines with income. There is substantial evidence of large impacts of college quality on college
completion (Cohodes and Goodman 2014; Bound, Lovenheim and Turner 2010), time to degree
(Bound, Lovenheim and Turner 2012), and subsequent earnings in the labor market (Andrews,
Li and Lovenheim forthcoming; Hoekstra 2009; Black and Smith 2006, 2004; Brewer, Eide and
Ehrenberg 1999).2 A representative estimate from Hoekstra (2009) shows that attending the
public flagship university leads to a 24% increase in earnings. Hence, differences in college
quality between low-income and high-income students could significantly affect both collegiate
attainment and earnings gaps.
In order to develop policies to address the gaps in postsecondary investment that exist across
the income distribution, it first is important to understand why they are present. There are
five main explanations for why students from low-income households tend to graduate from
college in general, and from more elite colleges in particular, at lower rates. First, families
with fewer resources at the time of college usually have fewer resources with which to invest
in a child throughout his or her life. These resource differences develop into differences in
academic preparation for college during students’ teenage years (Cameron and Taber 2004;
Carneiro and Heckman, 2002). Second, there is increasing evidence that low-income students
face considerable information gaps that often preclude them from applying to and enrolling in
more selective schools, even when they are academically qualified and would pay little to nothing
in out-of-pocket costs (Hoxby and Avery 2013; Hoxby and Turner 2013). A third explanation
is that low-income students are affected by both academic and social “mismatch” when they
enroll in higher-quality schools. On average, such students have worse academic preparation
for college and often are not part of the dominant cultural majority, particularly at more elite
postsecondary institutions (Aucejo, Arcidiacono and Hotz 2013; Arcidiacono and Koedel, 2014;
Arcidiacono et al., 2011; Dillon and Smith 2013). Fourth, the complexity of the financial aid
1 This is not just a reflection of the differences in enrollment. Among those who enroll in any college, 3.7% of low-income students
enroll in a public flagship university, and 18.4% of high income students enroll in this school type.
2 On the other hand Dale and Krueger (2013, 2002) find little impact of college quality on earnings.
2
application may prevent students from applying for aid, and thus attending more expensive
colleges (Dynarski and Scott-Clayton, 2013, 2008, 2006; Bettinger, et al., 2012). Finally, lower
family resources may prevent families from investing in a higher-quality school (Lovenheim and
Reynolds 2013).
The many hurdles faced by low-income students in the higher education system suggest that
programs addressing several of these barriers simultaneously might be particularly effective at
supporting postsecondary education among students from low-income backgrounds at highquality universities. In this paper, we present the first analysis in the literature of a set of
interventions in Texas aimed at addressing this array of disadvantages faced by low-income
students. The Longhorn Opportunity Scholarship (LOS) program at the University of Texas at
Austin (UT Austin) and the Century Scholars (CS) program at Texas A&M University – College
Station, which are the two flagship schools of the Texas public higher education system, began
in 1999 and 2000, respectively.3 The programs targeted high schools that served low-income
students and traditionally sent few students to these institutions. Together, the LOS and CS
programs were implemented in 110 high schools in Texas. While entirely independent, the
programs offer a similar suite of interventions that attempt to address the set of disadvantages
faced by low-income students in the higher education system. First, program participants are
provided scholarships to help alleviate financial strain.4 Second, the programs include multiple
support services for students once enrolled. Finally, the programs contain extensive outreach,
with students going back to their high schools to share their experiences with prospective college
students and university staff providing information sessions. This outreach and recruitment of
students from low-income high schools helps overcome information barriers that may preclude
students from these schools from applying to and enrolling in an elite postsecondary school
(Hoxby and Turner 2013). They also have the potential to generate “spillover” effects by
inducing students in targeted schools who are not offered scholarships to attend the flagships or
other higher quality institutions. Thus, the CS and LOS programs are designed to address the
multitude of difficulties that students from low-income backgrounds can face at very selective
3 Details on the Century Scholars program can be found at https://scholarships.tamu.edu/Scholarship-Programs/Century-Scholars.
The Longhorn Scholars Program has since been discontinued though a description can be found in internet archives at
https://web.archive.org/web/20030622194253/http://www.utexas.edu/student/finaid/scholarships/los_index.html.
4 For example, CS scholars currently receive $5,000 per year for four years. Assuming scholarship amounts did not change, this
covered most of the $5,639 cost for tuition and fees in 2004. Similarly LOS scholars in 2002 received $4,000 per year from the
program. Tuition at UT-Austin in 2005 was $7,286.
3
universities: lack of information about college quality, lower academic preparation for college,
and lower financial resources.
We use administrative data from the State of Texas that allows us to link K-12 education records with higher education enrollment and performance information as well as earnings
records from the Texas unemployment insurance system. Using this data we exploit the timing
differences in the roll-out of the LOS/CS programs to identify their effects on higher education outcomes and post-college earnings. Because these programs were targeted towards highperforming students, we first generate a performance index using the extensive set of high school
test score information we have about each student. We then estimate difference-in-difference
models in which we compare changes in outcomes among high-ability students according to
their high school academic performance index in treated schools to those with the same class
rank based on achievement in untreated schools when the LOS/CS programs are implemented.
The main identification assumption in these models is that the trends in enrollment patterns
and outcomes among high-achieving students would have been the same in treated and untreated high schools absent the programs. This assumption may be strong due to the fact that
the treated schools are highly selected. In order to make this assumption more credible, we
construct a “trimmed common support” group using the rich information we have about the
demographics and college-sending patterns of each high school in Texas prior to 1999. Thus, our
analysis sample is comprised of the set of schools that are more observationally-similar across
the treatment and control groups than would be the case if we used all high schools in Texas.
This is a feasible identification strategy because much of the targeting for these programs was
based on geography. We also show evidence of common trends in flagship enrollment prior to
the treatments, and we find little evidence of demographic shifts among students due to the
treatments.
The results of our analysis suggest that the LOS and CS programs had large effects on
the likelihood students enrolled in a flagship, with the effects somewhat larger for the LOS
program and UT-Austin enrollment. Enrollment at UT-Austin increases by 58% and Texas
A&M enrollment increases by 49% among high achieving (top 30% of HS class in achievement)
college attendees from treated schools relative to those in observationally similar untreated
4
schools. These enrollment increases came from both reduced enrollment at two-year schools
and as less-selective four-year universities, which suggests these treatments increased college
quality substantially for students. Further, we find evidence that many students switched from
two-year to four-year schools other than UT-Austin and TAMU, suggesting there are spillover
effects to students who did not attend the flagships. Notably, we find little impact on the
likelihood of enrolling in any public school in Texas.
Students who enroll in the LOS program appear to benefit considerably from the program.
Because of the existence of spillovers, we focus on intent-to-treat estimates of the effect of being
in a high school treated by the LOS or CS programs. Further, given the lack of impacts on
overall attendance in college, we focus on students who attend a two-year or four-year public
college, though we also include estimates for all high school graduates. Exposure to the LOS
program increased the likelihood of high-achieving college attendees graduating with a fouryear degree in six years by 3.5 percentage points. We also show that the LOS program did
not lead high achieving students to major in less-technical subjects: in particular, there is no
change in STEM majoring. These findings suggest that the extra academic support services
were sufficient to overcome any academic mismatch effects. On the other hand, when we look
at low achieving (bottom 70%) students who are induced into higher quality colleges we do see
evidence of a shift from STEM and social science to liberal arts.
The LOS program also increased earnings substantially. High-achieving college attendees in
LOS high schools experienced a 4% increase in earnings 10+ years post-high school. Attributing
all of this increase to the LOS treatment indicates that the LOS program led to a doubling
of earnings. There were no significant earnings impacts on low-achievers. For CS students,
we do not find any statistically significant effects on 6-year college graduation, although the
point estimates are negative and there is some evidence of longer time to degree. We find
little evidence that the CS program increased earnings as well, but our estimates are rather
imprecise.
With the design of our analysis, we cannot determine how much of the impacts we find are
due to the change in school quality or the provision of supports and financial aid. Nonetheless,
it is important to point out that the LOS and CS programs do not affect admissions likelihood.
5
In particular, all students in the top 10% of their high school classes were eligible for enrollment
at UT and TAMU and at the targeted schools few students below these cutoffs qualified for
admission. As a result, we interpret our estimates as telling us whether a program that provides
a full package of academic and social supports for low-income students who otherwise would
not attend the flagships can successfully improve educational and labor market outcomes. Our
results suggest that, if the program induces these “marginal” students to attend, they are more
likely to succeed than at lower-quality institutions where they would, arguably, get less support.
This is a key finding for two reasons. First, it provides evidence that attending a higher-quality
school can generate substantial economic improvements for low-income and relatively highability students, provided they receive sufficient assistance while enrolled to offset their lack of
preparation. Second, the LOS and CS programs are easily replicable beyond Texas. The pillars
of the program - targeted recruitment, mentoring, special classes and financial assistance - are
within the tool sets of flagship institutions in any state.
2
The Longhorn Opportunity and Century Scholars Programs
2.1
Program Description
The Longhorn Opportunity Scholars and Century Scholars Programs were first implemented in
1999 and 2000, respectively, to increase enrollment rates for low-income and minority students
at UT-Austin and Texas A&M in the wake of the state’s affirmative action ban. The affirmative
action ban went into effect in 1997 and made it illegal for schools in the state to consider race
as a factor in admissions. The pre-existing affirmative action system was replaced by the Texas
Top 10% Rule in 1998, which stipulated that any student in the top 10% of his or her high
school class could attend any Texas public university.5 Post-1997, the vast majority of students
in UT-Austin and Texas A&M were admitted under this rule. As a result of the Top 10% rule,
during the period we study students ranked outside the top 10 percent of their class at high
schools serving low-income students were very unlikely to enroll in UT-Austin or Texas A&M.
Despite the fact that many students from low-income schools became eligible to attend Texas
A&M and UT-Austin under this rule, minority enrollment at these colleges fell dramatically.
5 The
ranking is determined by each school separately, but typically is based on student grade point average.
6
For example, Kain, O’Brien and Jargowsky (2005) show that after the affirmative action ban
went into effect, enrollment amongst African American students at UT-Austin declined by 68%,
and it declined by 72% at Texas A&M. Hispanic enrollment also dropped considerably, by 6%
and 25% at UT-Austin and Texas A&M, respectively. In response to these declines, the LOS
and CS programs were developed to try to recruit students from low-SES backgrounds to the
state flagships and to support their academic success whole enrolled. The LOS program initially
targeted 70 high schools that had high shares of low-income and minority students and few prior
applicants to UT-Austin. The CS program similarly targeted 40 low-income schools in Houston,
Dallas and San Antonio with few prior applicants to Texas A&M, eventually expanding to 70
schools, although there was some overlap between the two programs. Over 600 students are
admitted to Texas A&M and UT-Austin under these programs each year. Figures 1 and 2 show
the geographic distribution of LOS and CS schools in our estimation sample, respectively; they
are mostly located in the large urban centers in the state and hence the focus of these programs
is on the urban poor.
Though administered by different universities, the two programs are similar and are summed
up best by the Longhorn Opportunity Scholarship Brochure:
More than simply a scholarship, the program serves as the catalyst for the creation of a
comprehensive academic community development package with a three-fold aim: to identify
students who, through a variety of circumstances, might not have otherwise had either the
opportunity or the desire to attend The University; to deploy University resources to attract
them to Austin; and most importantly, to give these students the resources and attention that
will help them to succeed academically and ultimately become alumni of The University of Texas
at Austin.
while Texas A&M describes the century scholar program as follows:
The Century Scholars Program is more than just a monetary award; it offers students access to a first-rate education and programs that prepare students to become state, national, and
world leaders. The Century Scholars Program offers academic support and hands-on contact
with advisors, mentorships with faculty, freshman seminar course that focuses on academic and
personal success, campus involvement, community engagement, and civic responsibility, and opportunities to serve as a Century Scholars Ambassadors. Century Scholars receive professional
training in public speaking, interviewing, and presentation skills. The students may return to
their former high schools to share their experiences and help continue the Texas A&M tradition
of excellence. These skills are highly valued by any future employer, professional school, or
graduate program.6
For LOS, initially the plan was to only offer services to students who received financial
6 Reprinted
from https://scholarships.tamu.edu/century_scholars.aspx
7
support from the program which was restricted to a maximum number of scholarships per
high school. However, though they did not advertise the fact widely, other enrolled students
from targeted schools were eligible for the program services. Further, an administrator of the
LOS program informed us that students who did not qualify for LOS scholarship money directly
usually qualify for other scholarships. For CS, students from targeted high schools are admitted
to the program only via competitive admissions and also must maintain GPA minimums while
enrolled. Thus, due to the less than full compliance with the program and the fact that we do
not have access to information on whether a student was a program participant, our empirical
analysis below focuses on intention-to-treat (ITT) effects of attending a high school where
the program is offered. Nonetheless, in an attempt to provide some insight into the scale of
treatment-on-treated effects (TOTE) that assume that all students attending UT-Austin from
an LOS high school and all students attending Texas A&M from a CS high schools post program
implementation are treated. For reasons that we describe below, we consider these estimates
to likely reflect upper bounds of the true treatment effects.
Despite the differences in program implementation, there are consistent properties across
the programs that make them worth investigating together:
1. Most students are given additional financial aid if they enroll in the flagship school.
2. There is an active recruiting effort made at targeted high schools to try and overcome any
information barriers about cost, the likelihood of admission, and the value of attending a
higher-quality school that may have existed. Recruitment occurs through both university
staff and students who have gone through the programs. These students thus could address
issues pertaining to academic and social mismatch directly.
3. Once enrolled, the LOS and CS students are given access to dedicated academic support
services to tutoring, and mentoring. Furthermore, the LOS and CS programs establish formal enrolled student and alumni communities that offer support, guidance, and resources
to low-income students.
4. Additional services are unique to each program. CS provides professional training in job
acquisition skills while LOS provides access to sections of classes with low class sizes and
targeted instruction and free tutoring.
8
These interventions could influence several important postsecondary outcomes and earnings
in ambiguous directions that point to the need for an empirical analysis. In particular, we
might expect the LOS/CS programs to have a positive effect on student outcomes because
of the overall positive effects of college quality on educational attainment and earnings (e.g.,
Andrews, Li and Lovenheim, forthcoming; Bound, Lovenheim and Turner 2010; Hoekstra 2009;
Black and Smith 2004, 2006; Brewer, Eide and Ehrenberg 1999).7 The LOS/CS programs
should increase the likelihood that students enroll in UT-Austin and Texas A&M. Indeed,
in interviews with ten freshmen recipients of the Longhorn Opportunity Scholarship, Bhagat
(2004) finds that the financial, social, and academic supports offered by LOS were the primary
reasons that students selected the University of Texas at Austin, suggesting that the programs
had positive effects on enrolling. This is consistent with the evidence in Domina (2007) and
Andrews, Ranchhod and Sathy (2010) of higher flagship enrollment after the LOS/CS program
implementation among students in treated high schools. Outside of the flagships, the other
options for these students typically are worse in terms of the quality and resource levels of the
institution, including attending lower-quality four-year schools, attending a two-year college or
not attending college at all. Domina (2007) shows that while students in LOS/CS schools were
more likely to enroll in a flagship, they were just as likely to attend a non-selective four-year
school after the treatment was implemented. This finding suggests that the alternative for most
of these students is a two year school or no college at all. We examine the enrollment effects of
these programs directly below using richer and more comprehensive data on enrollment than
were used in this prior work. Our results suggest a more nuanced story that differs across LOS
and CS treatments.
Due to the increased flagship enrollment driven by the LOS and CS programs, they likely
led to a substantial increase in college quality for treated students. To provide some context,
USNews and World Report ranks UT-Austin as the 58th and TAMU as the 68th best national
universities. The next highest public institutions in the state are UT-Dallas ranked 145, Texas
Tech ranked 156, and University of Houston at 186. Table 1 provides information on selectivity
and resources of Texas public institutions. The table compares University of Texas at Austin
7 Another potential mechanism is that increased financial support provided by the programs may help students progress through
the higher education system by relaxing credit constraints. However there is very little evidence that credit constraints or financial
aid have more than a modest impact on students’ paths through college (e.g., Johnson 2013; Stinebrickner and Stinebrickner 2008;
Bettinger 2004).
9
and Texas A&M to “emerging research universities” (ERUs) and other four-year schools. The
ERU designation is for institutions that are eligible for a special pool of funds for increasing
research output. These are sometimes called “Tier1” schools as part of the goal of the program
is to increase the schools’ research and academic reputations to the top tier of public universities
in the US. For our purposes, this is a useful distinction as it provides a “second tier” of public
institutions below the flagships but with better resources than other institutions. This group
includes University of Texas at Arlington, UT at El Paso, UT at Dallas, UT at San Antonio,
Texas Tech University and the University of Houston. The means in the table show that
both flagships are substantially more selective than the ERUs and other 4-year institutions as
measured by SAT scores of incoming students. The flagships also spend substantially more
per-student, have lower student-faculty ratios, higher graduation rates and higher retention
rates.
The ambiguity in predicted impacts of the programs arises because of potential tension between overall college quality effects and the potential for academic “mismatch” that can occur
when students of lower academic preparation are brought into a more demanding educational
environment.8 The students affected by the LOS and CS programs tend to be high-achievers
in their high schools, but because they come from low-income schools they still may be underprepared for the rigors of a flagship university. If the LOS/CS programs induce students to enroll
in schools in which they are mismatched, they could lower these students’ degree attainment,
persistence, and future earnings. They also could shift these students to easier, potentially
less lucrative majors. Nonetheless, the LOS and CS programs provide a system of social and
academic supports that potentially mitigate the experience of mismatch. The programs provide
mentors from the respective universities as well as student mentors who are scholarship recipients, and who thus have similar backgrounds to the prospective students. These mentors could
have eased the transition from high school to college for scholarship recipients. The academic
supports – smaller classes and tutoring – could compensate for any deficits in prior academic
preparation as well.
As a result of these conflicting theoretical impacts, a priori, it is not possible to determine the
net effect of these targeted recruitment programs. The success or failure of these programs must
8 See
Arcidiacono and Lovenheim (2015) for an overview of the “quality-fit” tradeoff in higher education.
10
be determined empirically, and the fact that the theoretical predictions are ambiguous makes it
critical for policy to examine empirically their effects on student educational and labor market
outcomes in order to determine whether or not the LOS/CS models are an appropriate way to
stimulate higher attainment rates among low-income students at elite colleges and universities.
These arguments underscore the importance of conducting a rigorous analysis that can identify
the effects of these targeted recruitment programs on students, which is what this analysis aims
to do.
2.2
Prior Literature
While, to our knowledge, no prior work exists that examines the impact of this type of multifaceted treatment aimed at addressing the multiple disadvantages faced by students from
low-income backgrounds at selective higher education institutions, there are several important
studies that have examined programs that contain individual components of the CS and LOS
treatments. In particular, prior work has examined the impacts of college outreach programs
and financial aid, with very little research being done on targeted college services. An important
contribution of our analysis stems from the fact that it may not be enough to merely address
one of the disadvantages faced by low-income students. Instead, to increase the postsecondary
attainment of such students, particularly at highly-selective schools, it may be necessary to
provide multiple targeted interventions simultaneously. Our study will be the first to provide
evidence on this type of broad intervention for a low-income and heavily minority population.
Previous research on college outreach programs has not found strong evidence they increase
student academic outcomes. Using National Education Longitudinal Study of 1988 (NELS:88)
data, Domina (2009) studies the effect of being exposed to a college outreach program that
provides information on the college application process and, in some cases, tutoring support and
college counseling services for high school students. Domina reports that about 5% of students
in the NELS:88 sample are exposed to such a program. Using propensity score matching
techniques, he finds little evidence that exposure to an outreach program influences high school
achievement or college enrollment. In a randomized controlled trial of Upward Bound, Myers
et al. (2004) find largely the same results, except for a positive four-year college enrollment
11
effect.
These studies do not examine the impact on college quality other than the four-year/2-year
margin. However, a major effect of the type of college outreach embedded in the CS/LOS
programs might be to influence students to attend a flagship rather than a non-flagship school.
There is some evidence that college outreach can positively influence the quality of schools to
which students apply and enroll. Hoxby and Turner (2013) conduct a randomized controlled
trial in which they send detailed information to high-achieving, low-income students throughout
the United States on college enrollment strategies as well as information about selective schools
and their likelihood of admission. They also include application fee waivers. Their findings
suggest that simply providing these high-achieving, low-income students with information about
their probabilities of admission to different tiers of schools and expected costs has significant
effects on the types of colleges and universities to which these students apply and attend. The
LOS and CS programs provide similar information and recruiting techniques, and they thus
could have large effects on the school choices made by students in the targeted high schools.
We note, though, that our study goes beyond Hoxby and Turner (2013) as we are able to
look at the long-term outcomes of interventions such as theirs that induce high ability, low
income students to attend higher quality schools, provided such an intervention is combined
with support services.
Our proposed research also relates to a body of work examining the effect of financial aid
on student collegiate choices and outcomes. Evidence from state merit aid programs that offer
free or highly-reduced tuition to in-state students who attend a public institution suggest these
programs are successful at altering the college enrollment decisions of high-achieving students
(Cohodes and Goodman 2014; Cornwell, Mustard and Sridhar 2006; Dynarski 2000). However,
these programs do not tend to increase students’ academic performance in college and even
may reduce it because they induce many students to enroll in lower-resource schools than they
otherwise would have (Cohodes and Goodman 2014; Fitzpatrick and Jones 2012; Sjoquist and
Winters 2012).
Importantly, the LOS and CS programs should have the opposite college quality effect to
what has been found in the merit aid literature. The likely alternative for these students is a less-
12
selective and lower-resource state university, community college or no college at all.9 UT-Austin
and Texas A&M-College Station have much higher per-student expenditures, lower studentfaculty ratios and significantly higher 6-year graduation rates. In addition, both flagships have
student bodies with higher measured pre-collegiate academic ability relative to other public
colleges and universities in Texas, as measured by the SAT score. Any resulting peer effects,
therefore, may play a role in driving the education differences across these schools and could
have a positive impact on LOS/CS students (Stinebrickner and Stinebrickner 2006; Zimmerman
2003; Sacerdote 2001).
This project also relates to a sizable body of work that has been done on the Texas Top 10%
plan. The Top 10% plan was implemented in 1998 as an alternative to affirmative action. It gave
automatic admission to any student in the top 10% of his or her high school class to any public
college or university in Texas. There is a large literature exploring the effect of the Texas
Top 10% plan on enrollment and completion outcomes, especially among minority students.
This research tends to find that the Texas Top 10% plan increases enrollment among highachieving students at flagship schools (Daugherty, Martorell and McFarlin forthcoming; Niu
and Tienda 2010; Domina 2007;), especially those who were in high schools that traditionally
did not send many students to these schools (Long and Tienda 2008; Domina 2007). The effects
on completion are more ambiguous, with some studies finding a negative effect (Cortes 2010)
and some finding no effect (Daugherty, Martorell and McFarlin forthcoming).
While our proposed analysis studies a set of interventions that is separate from the Texas
Top 10% law, this admission policy change provides an important backdrop for our study and
likely independently influenced enrollment choices among many of the students in treated high
schools. We discuss in Section 4 how this policy affects our identification strategy.
3
Data
The data we use in this study come from three sources: administrative data from the Texas
Education Agency (TEA), administrative data from the Texas Higher Education Coordinating
Board (THECB), and quarterly earnings data from the Texas Workforce Commission (TWC).
9 While it is possible that some students would have attended private or out-of-state schools, we do not find evidence of this in
our analysis.
13
The data are housed at the Texas Schools Project, a University of Texas at Dallas Education
Research Center (ERC). These data allow one to follow a Texas student from Pre-Kindergarten
through college and into the workforce, provided individuals remain in Texas. We discuss each
of these data sets in turn.10
Beginning in 1992, the TEA began collecting administrative data on all students enrolled
in public schools in Texas. These data contain students’ grade, the school in which he or she
is enrolled, scores from state standardized tests, and a host of demographic and educational
characteristics such as race/ethnicity, gender, special education status, Title 1 status, whether
the student is eligible for free or reduced-price lunch, whether the student is at risk of dropping
out, and enrollment in gifted and talented programs. The test score data we use are from
the 11th grade Texas Assessment of Academic Skills (TAAS) exams for reading, writing and
mathematics. The TAAS exams are administered to all students in Texas, and they are “high
stakes” in the sense that students must achieve a passing score on them in order to graduate.
Because students can retake them, we use the lowest score for each student, which typically
corresponds to the score from the first time students take the exam. Although the TEA data
begin in 1992, in 1994 Texas redesigned the high school exams. We therefore exclude data from
before the 1996 graduating cohorts and use TEA data from the high school classes of 1996-2002.
The LOS/CS programs targeted only high-ability students at each school. Hence, we focus
our analysis on the top of the within-school achievement distribution, although we also report
estimates for lower-achieving students. We estimate the students’ academic ability as the
first principal component of a factor analysis model that includes 11th grade TAAS scores on
mathematics, reading and writing.11 As argued by Cunha and Heckman (2008) and Cunha,
Heckman and Schennach (2010), combining test scores in a factor model provides a stronger
proxy for student academic ability than using any one test score alone. Using this academic
ability factor, we rank students in his or her school-specific 11th grade cohort. Andrews, Li and
Lovenheim (forthcoming) present evidence that the within-high school rank on these exams is
highly correlated with whether one is admitted to a flagship university through the Top 10%
Rule,12 which is evidence that the relative rank on these exams is a good proxy for relative
10 The
data used in this project are virtually identical to those used in Andrews, Li and Lovenheim (forthcoming).
scale this index so it has a mean of 100 and a standard deviation of 10.
12 They show that admission through the Top 10% Rule is highly predictive of attending UT-Austin or Texas A&M, but conditional
11 We
14
academic rank in each high school.
Our higher education data from the THECB contain detailed information about college enrollment and key collegiate outcomes for all students who enroll in a public college or university
in the State of Texas. For these students, we observe the enrollment decision in every school in
each semester, major choice, the timing of all degrees received, and credits earned that we can
use to calculate GPAs. The quarterly earnings data from the TWC are from 2007-2012 and
contain earnings for every worker in Texas, with the exception of those working for the Federal
government or US Postal Service. A core difficulty with measuring earnings is that earnings
early in one’s career may not be indicative of permanent earnings (Haider and Solon 2006).
Because the LOS and CS programs are relatively recent, we are constrained in the length of the
post-high school time period over which we can observe earnings. We construct two measures of
earnings to provide insight into the role of timing. The first is average log quarterly earnings in
all quarters in which earnings are observed six or more years post-high school graduation. The
second uses all earnings observations that are at least ten years after high school graduation.
To construct our earnings measure, we take all quarterly earnings observations of $100 or
more that meet the time criteria13 and estimate a regression of log quarterly earnings on a set
of calendar year, quarter-of-year, and high school graduating cohort dummies. We then take
the residuals from this regression and calculate the person-specific mean log earnings residual.
These earnings residuals can be interpreted as individual-specific average earnings that have
been adjusted to account for year, quarter and cohort effects.
A core limitation of our data is that students only are followed if they attend college in
Texas and then work in the labor force in Texas post-graduation. The main concern is that the
LOS/CS programs induce students who would have attended an out-of-state or private school
to move to the in-state flagship.14 This would affect the interpretation of our estimates, as
it would appear that students are “upgrading” school quality due to the programs while in
on the relative rank on the TAAS test scores this variable loses its predictive power.
13 We also exclude all earnings that occur while an individual is enrolled in a Texas public graduate school as these earnings are
unlikely to be reflective of permanent earnings (we do not observe if the student is enrolled in a private or out-of-state graduate
school). Furthermore, we highlight that the data do not include zero earnings among workers. A worker-quarter observation only is
present if the worker has earnings in that quarter. Missing observations can be due to unemployment, labor force non-participation
or leaving the State of Texas. We do not include missing observations as zeros because we are unsure whether an individual has
left the state or is not working and residing in Texas. These sample restrictions and the way in which we construct our earnings
measures are very similar to the methods used by Andrews, Li and Lovenheim (2014; Forthcoming) with these data.
14 Daugherty, Martorell and McFarlin (forthcoming) show that the Top 10% Rule had just such an effect on student college-going
in a low-income district.
15
actuality they are just shifting from a similar out-of-state or private school to a public flagship
university. Of course, these students still would receive the increases services once enrolled
as well as the scholarship money, but any college quality effects would be muted. Thus, this
type of sorting likely would lead us to overstate the program impacts, especially if the students
induced to switch schools have higher innate ability, desire to attend college, and/or wealth
that would generate better college outcomes and earnings.
We address this potential bias in a few ways. First, we note that in the wider population
affected by LOS and CS, very few students attend out-of-state or private schools. Indeed, in
Texas overall only 18% of first-time 4-year college enrollees who were seniors in high school the
prior year attend an out-of-state school. While similar statistics for in-state private schools are
not available, only 12% of enrollment in Texas degree granting institutions is in private colleges.
Given the low income of students in LOS/CS schools, we would expect these numbers to be far
lower for our subpopulation of interest.
Second, and most importantly, while we cannot directly measure whether students would
have attended out-of-state or public schools, we can identify if the LOS and CS programs have
any impact on attending an in-state public school. Thus, the treatment effect is relative to
not attending college, attending a 2-year college, attending a private college, or attending an
out-of-state college. As we show below, we find little indication that treated students were
more likely to be observed in the data. Thus, for the programs to induce private/out-of-state
students to move to the flagships, there would have to be an offsetting increase in 2-year or
non-college attendance by other treated students, which is very unlikely.
In addition to sample selection that can occur at the college choice stage, there can be
selection post-college due to migration out of Texas. While it is uncommon for students to
move out-of-state after college, it occurs often enough to be of concern. According to the 20082012 American Communities Survey, 2% of individuals in Texas with a bachelor’s or higher
degree move to a different state each year. Assuming that this rate is cumulative, then up to
10% of college graduates may move out of state within 5 years. Of course, this measure is
unlikely to be cumulative: those in a cohort with the highest propensity to leave would have
already left in earlier years. Additionally, the figures do not break down whether a student
16
gets a degree from an in- or out-of-state school. We would expect the former to have a lower
leaving rate. Nonetheless, the figures also are not broken down by age, and so we might expect
younger people to be more likely to leave. We note as well that Andrews, Li and Lovenheim
(forthcoming) show that earnings of bachelor degree holders in Austin (home of UT-Austin)
and College Station (home of TAMU) who move out-of-state do not differ meaningfully from
those who remain in-state. Given this context, we operate primarily under the assumption
that any attrition in the earnings data is unrelated to whether one is treated by the CS/LOS
program. In support of this assumption, we show that the LOS/CS treatment is uncorrelated
with being missing from the earnings data.
4
Methodology
Our methodological approach to examining the effect of the LOS/CS programs on student college choice, academic outcomes and labor market earnings is to estimate difference-in-differences
models in which we compare changes in outcomes when students are treated to changes among
students in schools that are not treated. As discussed above, the LOS and CS programs are
most likely to affect higher-ability students. In our baseline model, we therefore restrict the
analysis to students who are in the top 30% of their high school class in a given year according
to the ability index discussed in Section 3. We focus on the top 30% of students rather than
the top 10% because our ability index is an imperfect proxy for class rank. The top 30% of
students accurately captures the groups that are potentially eligible for enrollment in a state
flagship; below the 70th percentile, there is virtually no flagship enrollment in treated or untreated schools in any of the years of our study. We also provide results for students in the
bottom 70% in order to examine whether the treatment effects are isolated to the students most
likely to be eligible to attend a flagship university or if there are spillovers to lower achieving
students.
We begin with a difference-in-difference model that allows us to identify intention-to-treat
effects of the LOS/CS programs:
Yijt = α + β1 LOS Schooljt + β2 CS Schooljt + Xijt Γ + φj + θt + εijt ,
17
(1)
where Yijt is an educational or labor market outcome of interest for student i from high school j
who is in 12th grade in year t, and X is a vector of individual characteristics such as high school
test scores, race, gender, and free/reduced price lunch status. The model also contains school
fixed effects (φj ) and year fixed effects (θt ). The main treatment variables, LOS School and
CS School, are indicators for whether the 12th grade cohort in school j and year t is eligible
for the LOS or CS programs, respectively.
In equation (1), the main parameters of interest are β1 and β2 , which show how outcomes
change among top 30% students in LOS/CS schools relative to top 30% students in untreated
schools when the programs are implemented. The main assumption under which β1 and β2
are identified is that the counterfactual trends in outcomes among schools not receiving the
treatment are the same as those among the treated schools. This identification assumption is
potentially strong, especially since the programs are targeted at low-income schools that could
have substantially different trends than non-LOS/CS schools absent the treatment.
In order to make this identification assumption more likely to hold, we restrict the analysis schools to the set of high schools schools with common support amongst key observable
characteristics that determine treatment, in particular low prior flagship enrollment and low
income levels. Using data from the 1997-1998 school year (which is before either program was
implemented but after implementation of the Top 10% rule), we estimate a probit regression of
the likelihood a high school becomes an LOS or CS school as a function of the quadratic polynomials in the following school-level characteristics: percent enrolling in UT-Austin or Texas
A&M, percent taking the SAT or ACT, percent scoring above either 24 on the ACT or 1120 on
the math and verbal sections of the SAT (“college ready”), percent economically disadvantaged,
percent black, and percent Hispanic. The first three variables account for under-representation
at the flagship by measuring how many students are potentially eligible to attend the flagships
and how many actually enroll. The last three variables account for the socioeconomic makeup
of students in the schools. We estimate this model separately for LOS and CS treatments, and
we use this model to calculate a propensity score that shows the likelihood a given high school
is treated by each program.
The probit regression estimates are shown in Table 2. For both LOS and CS schools, the
18
strongest explanatory factor is the racial composition: high schools with high black or Hispanic
populations are more likely to be treated. For the Longhorn Opportunity Scholars treatment,
the percent taking the SAT also is an important predictor of treatment. Economic disadvantage
rates appear to play a role as well, although this measure is only significant for CS schools.
Interestingly, despite the universities’ stated goal of targeting schools with low prior attendance
rates at UT and TAMU, enrollment does not have a statistically significant relationship with
treatment. Some of this lack of a statistically significant effect of prior enrollment rates could be
driven by the strong correlation between flagship enrollment and school socioeconomic status,
so we urge caution in interpreting this result. Indeed, unconditional on student characteristics,
these pre-treatment enrollment rates are strongly related to the likelihood of being selected to
be an LOS or CS school.
In order to generate a common support sample that is likely to exhibit similar counterfactual
trends, we drop all control schools with a predicted likelihood of treatment less than 0.0515
and all treated schools with a predicted treatment likelihood higher than the highest control
school. We construct this trimmed analysis sample separately for the LOS treatment and for
the CS treatment and then pool the two analytic samples together to estimate equation (1).
Thus, our trimmed common support sample is comprised of a set of schools that have broadly
similar likelihoods of being treated based on their observable characteristics. Figure 1 shows
the propensity score densities for treated and control schools by likelihood bin, separately for
UT-Austin (LOS) and Texas A&M (CS), respectively. In the figure, we have excluded the
large mass of control schools with propensity scores below 0.05 as they dominate the graph if
included. Ostensibly, we are excluding a large set of high schools that serve higher-SES students
and thus that have no probability of being selected for the LOS/CS treatments. As the figures
demonstrate, there also are several treated schools that have a predicted likelihood of treatment
that is greater than any control school. These schools are shown in green: they are excluded
from the main analysis because they are sufficiently different from any comparison school that
it makes the identification assumptions underlying our estimator more difficult to support. We
refer to the sample that excludes these very high and low treatment likelihood schools as the
“trimmed common support sample.”
15 No
treated school has a propensity score value below 0.05.
19
Table 3 provides summary statistics for Texas college attendees in the trimmed common
support sample by whether or not they attended an LOS/CS high school before and after
the programs began.16 In general, the table indicates that the sample trimming brings the
observable characteristics of the students in treated and comparison schools during the pretreatment period (the first two columns) in line with each other. For both programs the students
in each school type have similar test scores, economic status, college enrollment rates, flagship
enrollment rates, and at-risk status. The only key differences are that treated schools have
more black students and fewer Hispanic students, but the white shares are similar. Comparing
across columns, any level differences in observables across treated and control schools remain
relatively constant over time.
Table 4 provides more direct tests for whether the implementation of the programs is correlated with changes in school composition. In the table, we provide balance tests using equation
(1), in which we exclude the observable characteristics in X and use each observable shown in
the columns of the table as a dependent variable. In Panel A, we focus on college attendees
who were in the top 30% of their high school ability index, while in Panel B we examine top
30% high school graduates. Panels C and D show balance tests for the bottom 70% sample.
Our preferred sample is the top 30% of students restricted to college attendees, as this is the
group most likely on the margin of treatment. Among these students, there is scant evidence
that the observable characteristics of students change when the treatments are enacted. For
LOS, there is one coefficient that is significant at the 5% level, but it is very small, suggesting
a 0.7 of a percent increase in TAAS writing scores. Similarly, there is only one estimate for
the CS treatment that is statistically significant at the 5% level. It also is modestly sized and
suggests that CS treatment is associated with high schools becoming lower SES. Furthermore,
for both LOS and CS some of the race/ethnicity estimates are significant at the 10% level, but
they are small in magnitude and are not in a consistent direction. Estimates for the top 30%
high school graduates are similar and are inconsistent with large changes in the demographic
characteristics of schools surrounding treatment that would bias our results. The same conclusion can be drawn from the bottom 70% samples in Panels C and D. Nonetheless, we note that
16 As we discuss in Section 5, there is no effect of LOS or CS on college enrollment, which supports our use of the college enrollment
sample.
20
we control for these characteristics in our regressions.
Given the targeted nature of these programs, it is important to understand what drives
the assignment to the treatment conditional on the observables. Figures 2 and 3 show the
geographic distribution of LOS and CS schools, respectively, as well as the trimmed common
support schools. As the figures demonstrate, much of the treatment variation is geographic:
the LOS and CS programs were targeted towards urban high schools in the largest cities in
Texas. Thus, there are many observationally-equivalent schools that are not located in these
cities that comprise much of the control groups. There are some control schools in these cities
as well. However, they tend to be located outside the urban centers and reflect the fact that
these programs faced budget constraints that allowed them only to treat a subset of qualifying
schools. Figures 2 and 3 suggest that there is plausibly exogenous variation in treatment status
based on geography that allows us to identify β1 and β2 in equation (1). To take explicit
advantage of this, we provide robustness checks below that use only control schools outside
of treated areas that are categorically excluded from being treated, as they may form a more
credible control group.
The central conditions needed for identification are common to any difference-in-difference
model: outcomes in the treated and control schools must be trending similarly prior to treatment
and there must not be shocks in 1999-2002 that affected CS/LOS schools differently from the
control schools. Our trimmed common support sample makes these assumptions more likely to
hold, but it still is important to provide direct evidence on their validity. Thus, we estimate
event study models in which we interact indicators for whether a school will ever be treated
by the LOS or CS programs with each calendar year. This allows us to test explicitly for the
existence of differential pre-treatment trends in outcomes. We find no evidence such trends
exist, which supports our empirical strategy. It is more difficult to test for unobserved shocks
that differentially impact the treated high schools. Of particular concern is the imposition of
the Top 10% Rule in 1998. As a result of this rule, most admissions to the flagship schools were
from the top 10% of a class. Equation (1) is identified under the assumption that the top 30%
in the treated and control schools are similarly affected by the Top 10% Plan. This assumption
is made more palatable by the use of the trimmed common support sample, since both treated
21
and control schools serve low-SES students with low historical flagship enrollment rates (see
Table 3). However, our event study estimates also shed light on any bias from the Top 10%
Plan as this law went into effect in 1998 while the LOS/CS treatments were not rolled out until
1999-2000. We therefore should see effects in 1998 if the Top 10% Rule is driving our estimates,
but as shown below the time pattern of effects much more closely matches the timing of the
LOS/CS rollout than the Top 10% Plan implementation.
Equation (1) is designed to identify intent-to-treat (ITT) parameters. That is, β1 and β2 in
equation (1) shows the effect of being exposed to the LOS/CS intervention by being in a treated
high school (or by being a high-performing student in a treated high school). From a policy perspective, this is an extremely important parameter because universities cannot compel take-up.
In addition, there can be spillover effects onto students who do not receive a LOS/CS scholarship, particularly from the recruitment part of the programs. Thus, from the policymaker’s
standpoint, the ITT is the most relevant parameter. However, the treatment-on-the-treated
effect (TOTE) also is a policy parameter of high interest in this setting. This parameter shows
the effect of receiving the services associated with these programs on educational and labor market outcomes, which informs how these treatments impact outcomes among treated students.
We face a data limitation in calculating the TOTE driven by the fact that we do not observe
who received scholarships. As discussed in Section 2, according to discussions with an LOS
program administrator, all students from LOS schools who attended UT-Austin were enrolled
in the program. While most, but not all, of the LOS scholars received financial aid through the
program, they all had access to the supports and mentoring provided by LOS. Furthermore,
most students who did not receive scholarships through LOS were able to receive financial support from other programs at the university above and beyond their Federal grants. It therefore
is reasonable to consider all UT-Austin students from LOS high schools who attended after the
program was as treated since they receive at least some of the services under the LOS program.
For the CS treatment, it is less clear who received services because the program was much more
limited in scope. We proceed under the same assumption for these students, that all post-CS
program attendees from CS high schools are treated.
On it’s own this assumption would provide a lower bound on the TOTE for this intervention.
22
However, the potential for spillovers to students who do not attend the flagships complicates
this. In the presence of spillovers, some of the impacts on non-flagship students would be applied
to flagship attendees rather than non-attendees, thus increasing the estimate. Essentially, this
increases the numerator in the Wald estimator without increasing the denominator. Given
substantial evidence shown below below of non-trivial spillover effects in college enrollment
behavior, in order to interpret the TOTE estimates we assume that the upwards bias from
the increase in college quality for non-flagship attendees is larger than downwards bias from
attributing treatment to all flagship attendees from treated schools. This allows us to treat our
TOTE estimates as upper-bounds on the true treatment effects.
To estimate treatment effects, we use enrollment in UT for a student from an LOS school in
the post adoption period as being a direct measure of treatment and enrollment in Texas A&M
from a CS high school post adoption as being treated by the CS program. Since the choice to
enroll is endogenous, we employ a two-stage least squares (2SLS) model that instruments LOS
enrollment with whether the student attended an LOS high school and CS enrollment with
whether the student attended a CS high school. Specifically, we estimate
LOS studentijt =γ0 + γ1 LOS Schooljt + γ2 CS Schooljt + Xijt Ω + δj + νt + μijt
(2)
CS studentijt =λ0 + λ1 LOS Schooljt + λ2 CS Schooljt + Xijt Ω + δj + νt + μijt
(3)
Yijt =α + β1 LOS Studentijt + β2 CS Student
ijt + Xijt Γ + φj + θt + εijt ,
(4)
where LOS Studentijt equals one if the student attended an LOS high school after the school
was identified as LOS and subsequently enrolls in UT-Austin, CS studentijt is a similar indicator for enrolling in a Texas A&M from a CS high school after the CS program was implemented,
and all other variables are as previously defined in equation (1). The parameters β1 and β2
in equation (4) are the TOTE estimates that use the roll-out of the LOS and CS programs as
instruments for whether a student is treated by the program. The identification assumptions
in this model are virtually identical to those underlying equation (1), with the additional condition that γ1 and λ2 need to be strongly related to treatment. Because of the difficulties with
interpreting these estimates described above, we focus predominantly on the ITT results, but
23
we show the TOTE estimates as well to provide some context to the magnitudes of the ITT
results.
5
Results
5.1
Main Results
Estimates of equation (1) using college enrollment outcomes as the dependent variable are shown
in Table 5. In the table, each set of two estimates in a column is from a separate regression.
Panel A shows estimates for college attendees and Panel B shows estimates for high school
graduates. In both panels we provide results for the top 30% and bottom 70% separately.17
Our preferred (and most restrictive) sample is the top 30% college attendee sample, and thus
we will focus on results for that sample throughout most of the paper. All estimates shown in
Table 5 and throughout the remainder of the paper use the trimmed common support sample.
In Panel B of the first column of results, we provide estimates of the effect of the CS/LOS
treatments on attending a public college in Texas. Recall that one potential complication from
our study is that, since we only have data on students who attend public colleges in Texas, if
the programs induce students to enter the public university system from other places - such
as private schools, out-of-state schools, or not attending college at all - this could generate
a sample selection bias. In neither the top 30% sample nor the bottom 70% sample of high
school graduates is there any evidence of a change in enrollment in a public Texas 2-year or
4-year college or university due to the CS/LOS programs. The coefficients are small with small
standard errors, and they are not statistically significantly different from zero. These results
support our focus on the college attendee sample when we examine collegiate outcomes and
earnings.
Columns (2) and (3) of Table 5 provide estimates of the impact of attending an LOS or
CS high school on enrollment at a flagship. When restricting to the top 30% among college
attendees, we find an increase in attendance of 2.8 percentage points in UT-Austin due to LOS
exposure and an increase of 1.7 percentage points in Texas A&M enrollment due to CS exposure.
Relative to the sample means, these estimates imply an increase in UT-Austin enrollment of
17 Note that, since enrollment in UT and TAMU is part of the treatment, we cannot show TOTE impacts on whether the student
enrolls in college or in which type of school a student enrolls.
24
58% and an increase in Texas A&M enrollment of 49%. The effects in Panel B for high school
graduates are similarly-sized relative to the sample means. Importantly, the CS treatment did
not affect enrollment in UT-Austin, nor did LOS treatment affect enrollment in Texas A&M.
This result suggests these programs were not simply moving students across flagship schools,
and they are inconsistent with differential secular enrollment trends confounding our estimates,
as these would likely affect enrollment in both flagships. We also find at most very small effects
of LOS or CS treatment on flagship enrollment among the bottom 70% students, which is
expected because a very small proportion of these students could be admitted to such a school.
As discussed above, a core identification assumption embedded in equation (1) is that the
treatment and control schools are trending similarly prior to the treatment rollout. In order
to provide evidence in support of this assumption, Figures 4 and 5 show event study estimates
of enrolling in UT-Austin and Texas A&M for the top-30% college and high school samples,
respectively. Across all figures, there no evidence of a differential upward trend in UT-Austin
or Texas A&M enrollment prior to treatment. Indeed, there is some evidence that UT-Austin
enrollment was declining in the treated schools relative to the control schools before the LOS
program was implemented. In both samples, there is a clear increase in flagship enrollment after
1999 among students in treated schools when the LOS and CS programs first began that is not
predictable from pre-treatment relative trends. Furthermore, these estimates suggest that the
Top 10% Rule is not a serious confounder in this setup, as there is no apparent increase in 1998
(the first year of the Top 10% Rule). That is, any differential changes in enrollment between
treated and untreated schools start to occur in 2000 after LOS and CS were implemented, not
in 1998 when Texas Top 10% is implemented. Overall, Figures 4 and 5 are consistent with
the identification assumptions underlying our difference-in-difference approach of common pretreatment trends between treatment and control and the lack of shocks that differentially affect
one group when the treatment occurs.
Since the LOS/CS programs did not affect the extensive margin of college enrollment and
did not shift students across flagship schools, it is important to understand where the changes
in enrollment came from. The remainder of Table 5 explores this question. We combine nonflagship universities into 3 mutually exclusive sectors: emerging research universities,18 other
18 These
emerging research universities are listed in Section 3.
25
four-year universities, and community colleges.19 Although there is some variability across
samples, three general patterns emerge. First, much of the increase in UT enrollment for the
LOS treatment is driven by declines in two-year enrollment. Thus, it appears that the LOS
program takes students who would have enrolled at a local community college and induces
them to enroll at UT-Austin. As Table 1 indicates, this represents a dramatic increase in
college quality for these students. That the LOS treatment shifts students from a two-year
to a flagship school is a very important finding given the fact that these students were not
given admission help; they were eligible to attend UT-Austin before the LOS program was
implemented but chose not to. This finding is consistent with evidence from Hoxby and Avery
(2013) that low-income, high-achieving students systematically choose less-selective schools
than their higher-income counterparts and suggests that programs like the LOS scholarship
can successfully get these students to enroll in more-selective schools.
Second, the CS treatment increases flagship enrollment more at the expense of emerging
research enrollment than two-year enrollment. Thus, the CS treatment led to a smaller increase in college quality than the LOS treatment. The third pattern evident in Table 5 is
that there are spillovers from the LOS and CS programs to students who do not enroll in
flagships as enrollment in non-flagship four-year schools increases, again likely at the expense
of 2-year school enrollment. While unexpected, we believe this is a result of the recruitment
efforts that UT-Austin and Texas A&M made under these programs. These recruitment efforts
plausibly induced many students to attend a four-year rather than a two-year college, even
if they could not get into or chose not to go to a flagship. That enrollment in non-flagship
four-year schools increases more among the bottom 70% students reinforces this conclusion.
The increased four-year non-flagship enrollment suggests that TOTE estimates will overstate
the effect of treatment,per se, as it will attribute outcome changes from students that did not
receive an LOS or CS scholarship to those who did.
Thus far, our results indicate that students in LOS and CS schools experienced a substantial
increase in college quality by shifting from lower-resource public schools to UT-Austin and
Texas A&M. The prior literature on the educational returns to college quality suggest that
these interventions should lead to higher BA receipt (Cohodes and Goodman 2014; Bound,
19 A
very small number of students attend health science campuses that we do not separately identify in this analysis.
26
Lovenheim and Turner 2010). In Table 6, we examine how the LOS and CS programs affected
four- and six-year degree completion. The structure of the table is almost identical to that
of Table 5, except here we provide both ITT and TOTE estimates in each panel. We do not
present TOTE estimates for the bottom 70% sample because there is a very small first-stage
flagship enrollment effect among this group. First, we examine first-year GPA to see whether
students are performing better or worse when they attend a more-selective school. The effects
are of opposite sign across programs, with those coming from LOS high schools experiencing
an increase of 0.11 GPA points and GPAs among students from CS schools declining by 0.08
points. There is no evident effect among bottom 70% students, however.
The different effects of the LOS and CS programs on first-year GPA are similar to the differences in program effects on BA completion. In the top 30% sample among college attendees,
there is a large, statistically significant effect on six-year graduation of 3.5 percentage points.
This is a 10.4% increase relative to the mean for this group. In contrast, the CS treatment has
a negative effect on the four-year graduation rate that is substantially attenuated and statistically insignifiant by six years. Thus, the CS program leads to a delay in graduation and it may
also decrease graduation rates somewhat. However, both programs increase the likelihood that
students graduate from the respective flagship university. We hypothesize that the different
graduation effects across treatments relates to the scope of the two different programs - there
appear to be more substantial academic supports in LOS than in CS - as well as the fact that
the LOS program likely led to a much larger change in college quality than CS. However, we
lack the ability to test these mechanisms directly. Figures 6 and 7 present event study estimates
that are less precise but consistent with these results. Critically, there is no evidence in these
figures of differential pre-treatment trends that could bias the results in Table 6.
The TOTE estimates, particularly for the LOS treatment, show that the reduced form effect
on six-year graduation rates is enormous relative to the first stage. Being treated by the LOS
program increases the likelihood of BA receipt by 57.7 percentage points in the college attendee
sample and by almost 100 percentage points in the high school graduate sample. The estimate
close to one is an outgrowth of our use of a linear probability model for ease of interpretation
and the fact that these are likely upper bounds. Consistent with the ITT results, there is
27
no evident increase, and perhaps even a decline, in six-year graduation among those treated
by the CS program. Across samples, the first stages are universally well-powered, with firststage F-statistics well above 10.20 Overall, the results in Table 6 suggest that when these lowincome students from disadvantaged backgrounds are induced to attend a high-quality flagship,
their educational outcomes improve or, at least in the case of TAMU, do not significantly
worsen. This finding runs counter to what one would expect if the students are academically
mismatched to the more demanding educational environment and is consistent with there being
a large positive effect of college quality on the likelihood of completing college. It is important
to emphasize, though, that this is not a test of mismatch as the students attending these
colleges are also receiving enhanced academic services. One interpretation of these results that
is important for policy is that these academic services are more than sufficient to overcome any
academic mismatch faced by the treated students.
Another prediction of mismatch theory is that under-prepared students will gravitate to
easier majors when they are overmatched. In Table 7, we examine whether enrolling in the
CS or LOS programs induces students to alter their chosen course of study. We focus in
this table on the student’s “final major,” which is either the major at graduation or the last
observed major for students who do not graduate from a public Texas college by the end of our
sample period.21 The results for the top 30% sample in Table 7 are inconsistent with mismatch
effects: students are more likely to major in liberal arts and are less likely to major in “other”
subjects. This other category is comprised mostly of vocational and technical support majors,
and thus these major changes reflects the fact that students are switching out of two-year and
less-selective four-year schools. Importantly, there is no negative effect on STEM majoring,
and there is a positive STEM estimate for the CS treatment that is significant at the 10%
level. These are particularly important findings because of the growing evidence that mismatch
leads to students shifting to easier majors (Arcidiacono, Aucejo and Hotz 2013; Arcidiacono,
Aucejo and Spenner 2012). We find little evidence to support such mismatch effects here for
high achievers. On net, students’ major choice is not highly affected by the LOS/CS programs.
That students are not majoring in easier subjects but are attending more elite schools and
20 We do not show estimates of the effect of LOS on Texas A&M enrollment (γ in equation (2)) or the effect of CS on UT-Austin
2
enrollment (λ1 in equation (3) for the sake of brevity. These estimates are both close to zero and not statistically significant in all
regressions and are available from the authors upon request.
21 In results available upon request, we show that these patterns are similar for initial major).
28
graduating at higher rates suggests the LOS program in particular led to a large increase in
human capital accumulation. Nonetheless, we note that when we look at the bottom 70% the
spillover effects appear to induce some students who switch to higher quality colleges to move
away from STEM majors. This further highlights the potential for the extra academic supports
- which are not available to the bottom 70% sample - to offset mismatch.
The large returns to college quality (Andrews, Li and Lovenheim forthcoming; Hoekstra
2009; Black and Smith 2004, 2006; Brewer, Eide and Ehrenberg 1999) combined with the
suggestive evidence of larger returns to more technical majors (Andrews, Li and Lovenheim,
forthcoming; Altonji, Blom and Meghir 2012; Arcidiacono 2004) suggest that the LOS and CS
interventions should raise earnings after college. In Table 8, we examine the effect of these
programs on earnings, using the adjusted log quarterly earnings measures discussed in Section
3. In the first two columns, we examine whether being in an LOS or CS high school affected
the likelihood that one appears in the earnings data. The estimates are close to zero and are
precisely estimated, suggesting treatment does not cause a sample selection problem.
In the remaining columns of Table 8, we show both short-term and medium-term effects
using all earnings after 6 and 10 years post high school graduation. Arguably, given that many
students take more than 6 years to complete college and may attend graduate school, the 10+
year results should be more reflective of lifetime earnings. Both sets of estimates show large
effects of the LOS program on earnings. Being in an LOS high school increases earnings after
6 years by 3.9% and after 10 years by 4.4%. These estimates translate into very large TOTE
effects: converting the estimates among the top 30% of college attendees to percent,22 being
treated by the LOS program increases earnings after 6 years by up to 93% and earnings after 10
years by as much as 111%. The large size of these estimates is consistent with the dramatic shift
in college quality and, for the LOS treatment, the sizable increase in the likelihood of graduating
from college. The results among top-30% high school graduates are qualitatively similar but
are smaller and less precise. This occurs because there is far more earnings variance among
the high school sample, and the proportion of students who are on the margin of treatment is
smaller. We therefore favor the college attendee sample of high ability students. Nonetheless,
despite the shift towards 4-year institutions, we find no evidence of an earnings effect among
22 Recall
that the percent effect from a log model is given by eβ − 1.
29
the bottom 70% samples. Further, in contrast to the LOS estimates, there is little evidence in
Table 8 that the CS program increased earnings. The point estimates are small and are not
significantly different from zero. However, they are not very precise, and thus we cannot rule
out that the CS intervention increased earnings. Overall, these results indicate that the LOS
program had very large, positive effects on the long-run labor market outcomes of the targeted
low-SES students while the effects of the CS program are less clear.
5.2
Geography Robustness Check
As shown in Figures 1 and 2, the treatment designation had a strong geographic component.
It therefore is possible that the untreated schools outside of the treated areas form a better
control group, since they are ineligible for treatment. That is, control schools in treated areas
may be more problematic because in those areas UT and TAMU made decisions about which
schools to treat. It is unlikely that the flagships made decisions about which cities to treat,
however, based on the characteristics of schools outside of the treated areas.
In Table 9, we present estimates of equation (1) that exclude all control schools in the same
county as a treated school. For the sake of brevity, we only show ITT estimates for the top
30% sample.23 The results are very similar in sign to those shown in the previous section.
There is no selection into college due to the treatments, and there is a clear increase in flagship
enrollment. The LOS program leads to a similar-sized 6-year BA attainment increase as in
Table 6. For the CS treatment, Table 9 shows a positive 6-year BA attainment effect for the
CS treatment, although it only is significant at the 10% level in Panel B. Consequently, there
now is evidence of a large, positive earnings effect stemming from the CS program, however
the estimates are quite imprecise. Similarly, the earnings estimates for the LOS treatment are
positive and large in Panel A, but they are not statistically different from zero at conventional
levels. Overall, these results show that the use of control schools in the same geographic areas
as the treated schools is not driving our results and likely results in conservative estimates for
the CS treatment.
23 Estimates
for the bottom 70% sample and 2SLS estimates are available from the authors upon request.
30
5.3
Estimates by Gender
Because men and women attend college at different rates and have different labor force participation rates, it is instructive to examine effects by gender. Table 10 shows ITT estimates for
men and women separately for the top 30% college attendee sample.24 The LOS and CS treatments increased flagship enrollment among both men and women, although female enrollment
at Texas A&M did not exhibit a statistically significant increase due to the CS treatment. Men
were more responsive to the CS program and women to the LOS program in their enrollment
behavior. Both groups exhibited a decline in two-year enrollment from the LOS treatment, although for women it was not statistically significant at conventional levels. The LOS program
also increased 6-year BA attainment among men and women, with a somewhat larger effect for
men. The CS treatment had a negative effect on attainment for men and no effect for women.
The starkest differences in program effects come when examining earnings. Among men, the
LOS treatment increased earnings substantially, with an ITT effect of over 7% that is statistically significant at the 5% level. There is no CS effect. In contrast, the LOS program estimates
for earnings among women are much smaller and are not statistically differentiable from zero.
This is despite the fact that the UT-Austin enrollment effect is much larger among women than
men. Hence, even though women attended UT-Austin at higher rates and were more likely to
obtain BA degrees, the earnings effects of the LOS program are concentrated amongst men.
There is some evidence that women in CS high schools experienced earnings increases as well,
with positive coefficients that are similar in size to the LOS estimates. However, as with the
LOS results, the estimates are not statistically significantly different from zero. Overall, these
results show that the earnings effects were most prevalent among men for the LOS program,
although both men and women experienced increases in college quality and BA attainment
rates.
6
Conclusion
Persistent increases in the college wage premium combined with sluggish growth in collegiate
attainment, particularly among students from low-income backgrounds, make it of first-order
24 Results for the high school sample and the bottom 70% college attendee sample, as well as TOTE estimates, are available from
the authors upon request.
31
importance to understand what policies can reduce attainment gaps in higher education across
the socioeconomic distribution. Given the evidence of the educational and labor market returns
to college quality as well as the low enrollment rates among low-income students at elite schools,
policies designed to raise enrollment rates of disadvantaged students at high-quality colleges
have the potential to reduce these disparities. We study two examples of such policies in Texas,
the Longhorn Opportunity and Century Scholars programs, which were designed to address
the multitude of disadvantages faced by low-income students in higher education: information,
tuition subsidies, and academic support once enrolled. These programs were targeted to schools
that served large numbers of low-income students and that did not historically send many
students to UT-Austin or Texas A&M.
We combine the timing of the implementation of the LOS and CS programs with detailed
administrative data from K-12 records, higher education records and earnings as long as workers
remain in Texas and attend a public university. We implement a set of difference-in-difference
estimators using a trimmed common support sample of treated and comparison schools that
compare how the enrollment behavior, educational outcomes and earnings of high-ability students change when the programs are implemented in targeted high schools in 1999 and 2000.
Our estimates suggest that these types of bundled interventions can generate better outcomes
among targeted students. Both the LOS and CS programs induced many students to enroll in
UT-Austin and Texas A&M instead of lower-resource four-year and two-year institutions. This
shift towards the flagship provided a large quality upgrade relative to the schools the students
would have attended in the absence of the program. High-ability students affected by the LOS
program saw large and statistically significant increases in graduation likelihood, and we find no
evidence of academic mismatch in the form of students switching to “easier” majors. We find
no effect of CS treatment on the likelihood of graduating from college, however. These changes
in college enrollment and collegiate outcomes led to large increases in earnings, especially for
the LOS treatment. College students from LOS high schools experienced a large increase in
earnings, and our upper-bound treatment on the treated results indicate earnings may have
doubled for those who received the LOS treatment. The CS program, however, did not increase
earnings among those in affected high schools.
32
The results from this analysis suggest programs like the Longhorn Opportunity Scholarship hold much promise in promoting better postsecondary and labor market outcomes among
high-ability, low-income students. While the intervention includes a bundle of services, such
a bundle would be straightforward for other flagship universities in other states to replicate.
Furthermore, while it is unclear if the students treated by the program are actually “undermatched” for the state flagships, the results suggest that mismatch problems can be overcome
with sufficient support services. Crucially, programs like these are easily replicable and thus
have the potential to generate large improvements in college-going and labor market outcomes
for high ability, low-income students who would not typically attend a flagship institution. The
preliminary indication for the Century Scholar program, however, provides a cautious note as it
is not automatic that such a program will succeed in affecting postsecondary and labor market
outcomes. More work focusing on the specific ways in which these programs were implemented
and the implications for effectiveness would be of high value in order to better understand how
to structure these programs to maximize their positive effects on students.
33
References
[1] Altonji, Joseph G., Erica Blom and Costas Meghir. 2012. “Heterogeneity in Human Capital Investments:
High School Curriculum, College Major, and Careers.” NBER Working Paper No. 17985.
[2] Andrews, Rodney J., Jing Li and Michael F. Lovenheim. Forthcoming. “Quantile Treatment Effects of
College Quality on Earnings.” Journal of Human Resources.
[3] Andrews, Rodney J., Jing Li and Michael F. Lovenheim. 2104. “Heterogeneous Paths Through College:
Detailed Patterns and Relationships with Graduation and Earnings.” Economics of Education Review 42:
93-108.
[4] Andrews, Rodney J., Vimal Ranchhod, and Viji Sathy. 2010. “Estimating the Responsiveness of College
Applications to the Likelihood of Acceptance and Financial Assistance: Evidence from Texas.” Economics
of Education Review 29(1): 104-115.
[5] Arcidiacono, Peter. 2004. “Ability Sorting and the Returns to College Major.” Journal of Econometrics
121(1-2): 343-375.
[6] Arcidiacono, Peter, Esteban Aucejo, Hanming Fang, and Ken Spenner. 2011. “Does Affirmative Action Lead
to Mismatch? A New Test and Evidence.” Quantitative Economics 2(3): 303-333.
[7] Arcidiacono, Peter, Esteban Aucejo, and Ken Spenner. 2011. “What Happens After Enrollment? An Analysis
of the Time Path of Racial Differences in GPA and Major Choice.” IZA Journal of Labor Economics 1(5).
[8] Arcidiacono, Peter, Esteban M. Aucejo and V. Joseph Hotz. 2013. “University Differences in the Graduation
of Minorities in STEM Fields: Evidence from California.” NBER Working Paper No. 18799.
[9] Arcidiacono, Peter and Cory Koedel. 2014. “Race and College Success: Evidence from Missouri.” American
Economic Journal: Applied Economics 6(3): 20-57.
[10] Arcidiacono, Peter and Michael F. Lovenheim. Forthcoming. “Affirmative Action and the Quality-Fit Tradeoff.” Journal of Economic Literature.
[11] Autor, David H. “Skills, Education, and the Rise of Earnings Inequality Among the ‘Other 99 Percent’.”
Science 344(6186): 843-851.
[12] Autor, David H., Lawrence F. Katz and Melissa S. Kearney. 2008. “Trends in U.S. Wage Inequality:
Revising the Revisionists.” Review of Economics and Statistics 90(2): 300-323.
[13] Bailey, Martha J. and Susan M. Dynarski. 2011. “Inequality in Postsecondary Education.” In G.J. Duncan
and R.J. Murnane (eds.), Whither Opportunity? Rising Inequality, Schools, and Children’s Life Chances.
Russell Sage: New York, New York.
[14] Bettinger, Eric. “How Financial Aid Affects Persistence.” In C.M. Hoxby (ed.), College Choices: The
Economics of Where to Go, When to Go, and How to Pay for it. University of Chicago Press: Chicago.
[15] Bettinger, Eric P, Bridgett Terry Long, Philip Oreopoulos, and Lisa Sonbonmatsu. 2012. “The Role of Application Assistance and Information in College Decisions: Results from the H&R Block Fafsa Experiment.”
Quarterly Journal of Economics 127(3): 1205-1242.
[16] Bhagat, Geeta Srinivasan. 2004. “The Relationship between factors that Influence College Choice and
Persistence in Longhorn Opportunity Scholarship Recipients at The University of Texas at Austin.” Doctoral
Dissertation at the University of Texas at Austin.
[17] Black, Dan A. and Jeffrey A. Smith. 2004. “How Robust is the Evidence on the Effects of College Quality?
Evidence from Matching.” Journal of Econometrics 121(1-2): 99-124.
[18] Black, Dan A. and Jeffrey A. Smith. 2006. “Estimating the Returns to College Quality with Multiple
Proxies for Quality.” Journal of Labor Economics 24(3): 701-728.
[19] Bound, John, Michael F. Lovenheim and Sarah E. Turner. 2010. “Why Have College Completion Rates
Declined? An Analysis of Changing Student Preparation and Collegiate Resources.” American Economic
Journal: Applied Economics 2(3): 129-157.
34
[20] Bound, John, Michael F. Lovenheim and Sarah E. Turner. 2012. “Increasing Time to Baccalaureate Degree
in the United States.” Education Finance and Policy 7(4): 375-424.
[21] Brewer, Dominic J., Eric R. Eide and Ronald G. Ehrenberg. 1999. “Does It Pay to Attend an Elite Private
College? Cross-Cohort Evidence on the Effects of College Type on Earnings.” Journal of Human Resources
34(1): 104-123.
[22] Cameron, Stephen V. and Christopher Taber. 2004. “Estimation of Educational Borrowing Constraints
Using Returns to Schooling.” Journal of Political Economy 112(1): 132-182.
[23] Carneiro, Pedro and James J. Heckman. 2002. “The Evidence on Credit Constraints in Post-Secondary
Schooling.” The Economic Journal 112(482): 705-734.
[24] Cohodes, Sarah and Joshua S. Goodman. 2014. “Merit Aid, College Quality, and College Completion:
Massachusetts’ Adams Scholarship as an In-Kind Subsidy.” American Economic Journal: Applied Economics
6(4): 251-283.
[25] Cornwell, Christopher, David B. Mustard, and Deepa J. Sridhar. 2006. “The Enrollment Effects of MeritBased Financial Aid: Evidence from Georgia’s HOPE Program.” Journal of Labor Economics 24(4): 761-786.
[26] Cortes, Kalena E. “Do Bans on Affirmative Action Hurt Minority Students? Evidence from the Texas Top
10% Plan.” Economics of Education Review 29(6): 1110-1124.
[27] Cunha, Flavio and James J. Heckman. 2008. “Formulating, Identifying and Estimating the Technology of
Cognitive and Noncognitive Skill Formation.” Journal of Human Resources 43(4): 738-782.
[28] Cunha, Flavio, James J. Heckman and Susanne M. Schennach. 2010. “Estimating the Technology of Cognitive and Noncognitive Skill Formation.” Econometrica 78(3): 883-931.
[29] Dale, Stacey Berg and Alan B. Krueger, 2014. “Estimating the Return to College Selectivity over the
Career Using Administrative Earnings Data.” Journal of Human Resources 49(2): 323-358.
[30] Dale, Stacey Berg and Alan B. Krueger, 2002. “Estimating the Payoff to Attending A More Selective
College: An Application of Selection on Observables and Unobservables.” Quarterly Journal of Economics
117(4): 1491-1527.
[31] Daugherty, Lindsay, Francisco Martorell and Isaac McFarlin, Jr. Forthcoming. “Percent Plans, Automatic
Admissions, and College Enrollment Outcomes.” IZA Journal of Labor Economics.
[32] Dillon, Eleanor Wiske and Jeffrey Andrew Smith. 2013. “The Determinants of Mismatch Between Students
and Colleges.” NBER Working Paper No. 19286.
[33] Domina, Thurston. 2007. “Higher Education Policy as Secondary School Reform: Texas Public High Schools
after Hopwood.” Education Evaluation and Policy Analysis 29(3): 200-217.
[34] Domina, Thurston. 2009. “What Works in College Outreach: Assessing Targeted and Schoolwide Interventions for Disadvantaged Students.” Education Evaluation and Policy Analysis 31(2): 127-152.
[35] Dynarski, Susan. 2000. “Hope for Whom? Financial Aid for the Middle Class and Its Impact on College
Attendance.” National Tax Journal 53(3): 629-661.
[36] Dynarski, Susan and Judith Scott-Clayton. 2013. “Financial Aid Policy: Lessons from Research.” Future
of Children May.
[37] Dynarski, Susan and Judith Scott-Clayton. 2008. “Complexity and Targeting in Federal Student Aid: A
Quantitative Analysis.” Tax Policy and the Economy 22: 109-150.
[38] Dynarski, Susan and Judith Scott-Clayton. 2006. “The Cost of Complexity in Student Financial Aid:
Lessons from Optimal Tax Theory and Behavioral Economics.” National Tax Journal 59(2): 319-356.
[39] Fitzpatrick, Maria D. and Damon Jones. 2012. “Higher Education, Merit-Based Scholarships and PostBaccalaureate Migration.” NBER Working Paper No. 18530.
[40] Goldin, Claudia and Lawrence F. Katz. 2007. “Long-Run Changes in the Wage Structure: Narrowing,
Widening, Polarizing.” Brookings Papers on Economic Activity 2007(2): 135-165.
35
[41] Hoekstra, Mark. 2009. “The Effect of Attending the Flagship State University on Earnings: A DiscontinuityBased Approach.” Review of Economics and Statistics 91(4): 717-724.
[42] Hoxby, Caroline and Christopher Avery. 2013. “The Missing ”One-Offs”: The Hidden Supply of HighAchieving, Low-Income Students.” Brookings Papers on Economic Activity. Spring: 1-50.
[43] Hoxby, Caroline and Sarah Turner. 2013. “Expanding College Opportunities for High-achieving, Low Income Students.” Stanford Institute for Economic Policy Research Discussion Paper 12-014.
[44] Johnson, Matthew T. 2013. “Borrowing Constraints, College Enrollment, and Delayed Entry.” Journal of
Labor Economics 31(4): 669-725.
[45] Kaine, John F., Daniel M. O’Brien and Paul A. Jargowsky. 2005. “Hopwood and the Top 10 Percent Law:
How They Have Affected the College Enrollment Decisions of Texas High School Graduates.” Report to the
Andrew W. Mellon Foundation: http://www.utdallas.edu/research/tsp-erc/pdf/wp kain 2005 hopwood top
10 percent.pdf.
[46] Long, Mark and Marta Tienda. 2008. “Winners and Losers: Changes in Texas University Admissions
Post-Hopwood.” Education Evaluation and Policy Analysis 30(3): 255-280.
[47] Lovenheim, Michael F. and C. Lockwood Reynolds. 2013. “The Effect of Housing Wealth on College Choice:
Evidence from the Housing Boom.” Journal of Human Resources 48(1): 1-35.
[48] Myers, David, Rob Olsen, Neil Seftor, Julie Young, and Christina Tuttle. 2004. The Impacts of Regular
Upward Bound: Results from the Third Follow-up Data Collection. Washington, DC: Mathematica Policy
Research.
[49] Niu, Sunny Xinchun and Marta Tienda. 2010. “The Impact of the Texas Top Ten Percent Law on College
Enrollment: A Regression Discontinuity Approach.” Journal of Policy Analysis and Management 29(1):
84-110.
[50] Sacerdote, Bruce. 2001. “Peer Effects With Random Assignment: Results For Dartmouth Roommates.”
Quarterly Journal of Economics 116(2): 681-704.
[51] Sjoquist, David L. and John V. Winters. 2012. “State Merit-based Financial Aid Programs and College
Attainment.” IZA Discussion Paper No. 6801.
[52] Stinebrickner, Ralph and Todd R. Stinebrickner. 2006. “What Can be Learned about Peer Effects Using
College Roommates? Evidence from New Survey Data and Students from Disadvantaged Backgrounds.”
Journal of Public Economics 90(8-9): 1435-1454.
[53] Stinebrickner, Ralph and Stinebrickner, Todd. 2008. “The Effect of Credit Constraints on the College
Drop-Out Decision: A Direct Approach Using a New Panel Study.” American Economic Review 98(5):
2163-2184.
[54] Zimmerman, David J. 2003. “Peer Effects in Academic Outcomes: Evidence from a Natural Experiment.”
Review of Economics and Statistics 85(1): 9-23.
36
Figure 1: Distribution of LOS and CS Treatment Probabilities by Treatment Status
−.3
−.2
Density
−.1
0
.1
.2
.3
Common Support for LOS
0
.2
.4
.6
Propensity Score
Untreated
Treated: Off support
.8
1
Treated: On support
−.3
−.2
Density
−.1
0
.1
.2
Common Support for CS
0
.2
.4
.6
Propensity Score
Untreated
Treated: Off support
37
.8
Treated: On support
1
Figure 2: UT Austin Longhorn Opportunity Scholars and Comparison Schools
38
Figure 3: Texas A&M Century Scholars and Comparison Schools
39
Figure 4: Flagship Enrollment Trends by Treatment Status - Top 30% College Attendees Sample
−.02
0
Enrollment Percentage
.02
.04
.06
.08
Panel A: UT Enrollment − LOS Treatment
1996
1997
1998
1999
Year
2000
2001
2002
.04
.02
0
−.02
Enrollment Percentage
.06
.08
Panel B: TAMU Enrollment − CS Treatment
1996
1997
1998
1999
Year
40
2000
2001
2002
Figure 5: Flagship Enrollment Trends by Treatment Status - Top 30% HS Graduates Sample
−.02
Enrollment Percentage
0
.02
.04
Panel A: UT Enrollment − LOS Treatment
1996
1997
1998
1999
Year
2000
2001
2002
.04
.02
0
−.02
Enrollment Percentage
.06
.08
Panel B: TAMU Enrollment − CS Treatment
1996
1997
1998
1999
Year
41
2000
2001
2002
Figure 6: 6-Year Bachelor Attainment Trends by Treatment Status - Top 30% College Attendees
Sample
−.1
Graduation Percentage
−.05
0
.05
.1
Panel A: Graduate in 6 Years − LOS Treatment
1996
1997
1998
1999
Year
2000
2001
2002
−.05
Graduation Percentage
0
.05
.1
Panel B: Graduate in 6 Years − CS Treatment
1996
1997
1998
1999
Year
42
2000
2001
2002
Figure 7: 6-Year Bachelor Attainment Trends by Treatment Status - Top 30% HS Graduates
Sample
−.04
Graduation Percentage
−.02
0
.02
Panel A: Graduate in 6 Years − LOS Treatment
1996
1997
1998
1999
Year
2000
2001
2002
−.04
−.02
Graduation Percentage
0
.02
.04
Panel B: Graduate in 6 Years − CS Treatment
1996
1997
1998
1999
Year
43
2000
2001
2002
Table 1: Average Characteristics of Public 4-Year Institutions in Texas
School Characteristic
Max USNews Ranking
Graduation Rate
Retention Rate
Avg Full Prof Salary
UG Student/Faculty FTE
Instr Exp per UG Student
Acad Support Exp per UG Student
Student Service Exp per UG Student
SAT Math 75th Percentile
SAT Reading 75th Percentile
Institutions
UT-Austin
Texas A&M
53
0.79
0.94
$137,871
14.0
$19,320
$5,633
$1,761
710
680
1
68
0.79
0.91
$128,367
17.0
$13,421
$3,853
$1,914
630
610
1
Emerging
Research
145
0.47
0.76
$122,131
22.6
$7,880
$2,865
$1,572
588
553
7
Other
4-Year
NA
0.37
0.64
$87,352
21.2
$6,491
$2,229
$1,387
519
537
21
Means from Integrated Postsecondary Education Data System (IPEDS) provided by the US Department of Education. Data is from 2013-14 except expenditure data which is from 2012-13 school year.
“Emerging research” universities are institutions declared by state of Texas to be eligible for special
funds to increase research activity. These include UT-Dallas, UT-Arlington, UT-San Antonio, UT-El
Paso, Texas Tech and University of Houston.
44
Table 2: Probit Regressions of LOS/CS Eligibility on
LOS/CS Determination Factors
HS Characteristic
in 1998
% Taking SAT
(% Taking SAT)2
% College Ready
(% College Ready)2
% Econ Disadv
(% Econ Disadv)2
% Black
(% Black)2
% Hispanic
(% Hispanic)2
% Enroll in UT
(% Enroll in UT)2
% Enroll in TAMU
(% Enroll in TAMU)2
Dependent Variable
HS is a UT
HS is a TAMU
Longhorn School Century School
0.212**
-0.026
(0.092)
(0.058)
-0.0023**
0.0002
(0.0009)
(0.0005)
-0.132
-0.039
(0.081)
(0.048)
0.0018
0.0011
(0.003)
(0.001)
0.089
0.085*
(0.057)
(0.047)
-0.0007
-0.0005
(0.000)
(0.000)
0.124***
0.110***
(0.044)
(0.032)
0.0001
0.0002
(0.000)
(0.000)
0.155***
0.223***
(0.053)
(0.052)
-0.00060
-0.00140***
(0.00043)
(0.00041)
0.242
(0.630)
-0.283
(0.249)
0.168
(0.233)
-0.020
(0.034)
Observations
949
949
Notes: “% College Ready” is the share of students in the graduating
class who scored above 1100 on the math and reading portions of the
SAT exam or 24 on the ACT exam.
45
Table 3: Summary Statistics for Trimmed Common-Support Sample
TAAS Writing
(% Correct)
TAAS Reading
(% Correct)
TAAS Math
(% Correct)
All College Attendees
1996 - 1998
1999 - 2002
Treated Schools Untreated Schools Treated Schools Untreated Schools
(i)
(ii)
(iii)
(iv)
30.8
31.4
32.7
32.8
(6.5)
(6.4)
(5.9)
(6.1)
36.8
37.6
39.0
39.3
(8.1)
(7.9)
(7.1)
(7.1)
40.6
42.1
45.7
46.5
(11.7)
(11.7)
(10.8)
(10.6)
White
Black
Hispanic
Gifted & Talented
At Risk
Male
Econ. Disadvantaged
Enroll in UT
Enroll in TAMU
Graduate in 6 Yrs.
0.08
0.30
0.59
0.07
0.46
0.41
0.47
0.01
0.01
0.16
0.14
0.10
0.74
0.10
0.51
0.44
0.52
0.01
0.01
0.21
0.06
0.29
0.62
0.16
0.47
0.44
0.60
0.02
0.02
0.20
0.11
0.11
0.75
0.11
0.48
0.45
0.57
0.02
0.01
0.24
Resid. Log Earnings
(10+ Years)
0.10
(0.78)
0.11
(0.80)
-0.13
(0.84)
-0.15
(0.84)
Observations
7,254
29,292
11,028
42,080
Enroll in College
0.52
0.56
All HS Grads
0.51
0.55
Notes: Authors’ tabulations using college attendees from the linked ERC-THECB data for the 1996-2002 high school
graduating cohorts.
46
47
0.196
(0.165)
-0.100
(0.156)
106.9
LOS
Sample Means
93.1
0.087
(0.296)
-0.379
(0.281)
94.5
29.2
(6)
Black
(7)
Hisp
53.7
0.324
(0.251)
-0.018
(0.236)
0.163
0.017*
(0.010)
-0.019
(0.012)
0.124
-0.008
(0.008)
0.018*
(0.010)
0.687
-0.007
(0.010)
0.022*
(0.012)
53.8
0.345
(0.240)
-0.086
(0.211)
0.178
0.017*
(0.009)
-0.014
(0.009)
0.132
-0.011
(0.006)
0.005
(0.007)
0.666
-0.004
(0.009)
0.020**
(0.009)
35.9
0.106
(0.212)
-0.261
(0.193)
40.2
-0.028
(0.447)
-0.000
(0.401)
0.089
0.020***
(0.005)
-0.012*
(0.007)
0.158
-0.010*
(0.006)
0.001
(0.006)
0.735
-0.011
(0.007)
0.016*
(0.009)
Panel C: Bottom 70% College Attendees (N=60,989)
44.1
0.002
(0.110)
-0.128
(0.100)
Panel B: Top 30% High School Graduates (N=61,800)
44.0
0.031
(0.115)
-0.082
(0.121)
0.044
0.007
(0.013)
0.016**
(0.007)
0.269
0.017
(0.034)
0.017
(0.023)
0.243
0.033
(0.034)
0.035
(0.022)
34.9
0.135
(0.196)
-0.433**
(0.177)
38.9
-0.058
(0.427)
-0.371
(0.397)
0.091
0.017***
(0.006)
-0.012*
(0.007)
0.170
-0.012**
(0.005)
-0.011**
(0.005)
0.726
-0.006
(0.007)
0.029***
(0.008)
(8)
G&T
0.039
0.010
(0.013)
0.009
(0.005)
Panel D: Bottom 70% High School Graduates (N=141,156)
0.070
(0.149)
-0.069
(0.162)
30.0
0.006
(0.172)
0.256
(0.183)
36.7
0.137*
(0.081)
-0.013
(0.090)
36.6
0.207**
(0.096)
0.033
(0.102)
(5)
White
Panel A: Top 30% College Attendees (N=28,665)
TAAS Raw Scores
Writing
Read
Math
(2)
(3)
(4)
0.605
-0.028
(0.031)
0.004
(0.032)
0.592
-0.038
(0.034)
0.006
(0.040)
0.254
0.004
(0.028)
-0.018
(0.031)
0.261
0.008
(0.030)
-0.036
(0.035)
(9)
At-Risk
0.465
-0.006
(0.006)
0.012*
(0.007)
0.437
0.001
(0.009)
0.015
(0.011)
0.455
0.015*
(0.009)
0.007
(0.010)
0.450
-0.006
(0.014)
0.015
(0.017)
(10)
Male
0.591
0.035**
(0.017)
0.042**
(0.018)
0.575
0.033*
(0.018)
0.036*
(0.020)
0.495
0.028*
(0.017)
0.044**
(0.019)
0.498
0.013
(0.018)
0.050**
(0.024)
Econ
Disadv
(11)
Notes: Authors’ estimation of equation (1) in the text using data for the 1996-2002 high school graduating cohorts, excluding all student characteristics and
using the variable listed in the column title as the dependent variable. Each group of two coefficient estimates in each column comes from the same regression.
Standard errors clustered at the high school level are in parentheses: ***, **, * indicate significance at the 1%, 5% and 10% levels, respectively.
Sample Means
CS
LOS
Sample Means
CS
LOS
CS
0.046
(0.316)
0.018
(0.291)
106.8
Sample Means
CS
0.243
(0.172)
-0.028
(0.169)
Achievment
Index
(1)
LOS
Dep. Var. →
Table 4: Balance Tests for Trimmed Common-Support Samples
48
LOS
Top 30%
0.515
(3)
Attend TAMU
0.003
-0.001
(0.001)
0.003***
(0.001)
0.035
-0.005
(0.005)
0.017**
(0.007)
0.001
0.001**
(0.000)
-0.000
(0.000)
0.026
0.023***
(0.004)
0.001
(0.004)
0.002
-0.000
(0.001)
0.001**
(0.001)
0.020
0.006*
(0.003)
0.009**
(0.004)
Panel B: High School Graduates
0.003
0.002**
(0.001)
0.000
(0.001)
0.048
0.028***
(0.007)
-0.002
(0.008)
Panel A: College Attendees
(2)
Attend UT
0.018
0.000
(0.002)
-0.008***
(0.002)
0.070
0.002
(0.006)
-0.013*
(0.007)
0.039
-0.002
(0.004)
-0.013***
(0.004)
0.139
0.011
(0.010)
-0.005
(0.012)
Attend Other
Research U
(4)
0.101
0.014***
(0.005)
0.005
(0.005)
0.159
0.034***
(0.010)
0.013
(0.010)
0.220
0.043***
(0.012)
0.021*
(0.012)
0.308
0.023
(0.017)
0.010
(0.015)
Attend Other
4 Yr
(5)
0.476
-0.024**
(0.010)
-0.005
(0.009)
0.450
-0.072***
(0.018)
-0.013
(0.018)
0.735
-0.043***
(0.012)
-0.011
(0.012)
0.468
-0.037
(0.020)
-0.000
(0.016)
(6)
Attend 2yr
Notes: Estimation of equation (1) in the text using the linked ERC-THECB data for the 1996-2002 high school graduating
cohorts. Each group of two coefficient estimates in each column comes from the same regression. All models include high
school and year fixed effects as well as the demographic, high school and test score controls discussed in Section 4 of the
text. Sample sizes for the top 30% college attendee and HS grad samples are 28,665 and 61,800, respectively. Sample sizes
for the bottom 30% college attendee and HS grad samples are 60,989 and 141,156, respectively. Note that sample means
do not necessarily sum to one as do not include health science campuses. Standard errors clustered at the high school level
are in parentheses: ***, **, * indicate significance at the 1%, 5% and 10% levels, respectively.
Sample Mean
CS
-0.011
(0.011)
-0.022**
(0.009)
LOS
Bottom 70%
0.007
(0.015)
-0.010
(0.018)
0.627
CS
LOS
-
Sample Mean
Top 30%
Sample Mean
CS
-
Bottom 70%
LOS
-
-
Attend Any
TX College
(1)
Sample Mean
CS
Treatment
HS Class
Achievement
Table 5: The Effect of Attending a Longhorn Opportunity or Century Scholar High School
on College Enrollment
49
LOS
Top 30%
-
0.033
141,156
0.005**
(0.002)
-0.004*
(0.002)
0.074
61,800
0.014***
(0.005)
-0.011*
(0.006)
0.064
60,989
-0.004
(0.005)
-0.002
(0.004)
0.127
28,665
0.000
(0.007)
-0.031***
(0.009)
Grad TX College
in 4 Years
Grad UT
in 6 Years
Grad TAMU
in 6 Years
0.003
60,989
0.002**
(0.001)
-0.001
(0.001)
0.035
28,665
0.019***
(0.005)
0.003
(0.007)
0.003
60,989
-0.000
(0.001)
0.002*
(0.001)
0.025
28,665
-0.004
(0.005)
0.008*
(0.005)
0.084
141,156
0.006*
(0.003)
-0.008**
(0.004)
0.200
61,800
0.030***
(0.008)
-0.007
(0.009)
0.001
141,156
0.000
(0.000)
-0.001
(0.000)
0.025
61,800
0.013***
(0.003)
-0.002
(0.004)
0.002
141,156
-0.001***
(0.000)
0.001*
(0.000)
0.020
61,800
-0.002
(0.003)
0.001
(0.003)
Panel B: High School Graduates
0.168
60,989
0.004
(0.008)
-0.011
(0.008)
0.336
28,665
0.035***
(0.011)
-0.019
(0.013)
Panel A: College Attendees
ITT Estimates
Grad TX College
in 6 Years
61,800
0.030***
(0.003)
0.022***
(0.002)
28,665
0.060***
(0.005)
0.052***
(0.006)
First
Stage
-
-
27,258
1.621***
(0.526)
-1.089*
(0.619)
0.074
61,800
0.459***
(0.174)
-0.381
(0.236)
0.127
28,665
-0.016
(0.127)
-0.612***
(0.181)
0.200
61,800
0.992***
(0.256)
-0.120
(0.366)
0.336
28,665
0.577***
(0.200)
-0.287
(0.250)
Upper Bound TOT Estimates
First Year Grad TX College
Grad TX College
GPA
in 4 Years
in 6 Years
Notes: Estimation of equations (1) and (2)-(4) in the text using the linked ERC-THECB data for the 1996-2002 high school graduating cohorts. Each group of two coefficient
estimates in each column comes from the same regression. All models include high school and year fixed effects as well as the demographic, high school and test score controls
discussed in Section 4 of the text. The first-stage estimates of the effect of CS treatment on UT enrollment and the effect of LOS on TAMU enrollment are not shown, although
they are included in the model. These estimates are close to and are not statistically significantly different from zero. Standard errors clustered at the high school level are in
parentheses: ***, **, * indicate significance at the 1%, 5% and 10% levels, respectively.
Sample Mean
Obs
CS
-
LOS
Bottom 70%
-
-
CS
LOS
52,400
Sample Mean
Obs
Top 30%
Sample Mean
Obs
CS
0.048
(0.029)
0.006
(0.022)
Bottom 70%
LOS
27,258
0.105***
(0.032)
-0.071**
(0.030)
First Year
GPA
Sample Mean
Obs
CS
Treatment
HS Class
Achievement
Table 6: The Effect of Longhorn Opportunity and Century Scholar Programs on College Graduation and First-Year GPA
50
STEM
(4)
Agriculture
(5)
0.233
0.068***
(0.026)
0.038**
(0.016)
0.096
-0.006
(0.006)
-0.019***
(0.006)
0.111
-0.006
(0.008)
-0.000
(0.010)
0.040
-0.018***
(0.006)
0.026**
(0.012)
0.060
0.003
(0.007)
0.023*
(0.013)
0.57
-0.014***
(0.005)
-0.006
(0.005)
0.136
-0.002
(0.010)
0.017*
(0.009)
0.001
-0.000
(0.001)
0.000
(0.000)
0.002
-0.000
(0.001)
0.002
(0.001)
0.013
-0.003
(0.002)
-0.001
(0.002)
0.022
0.000
(0.004)
0.013**
(0.005)
Communications
(6)
Attends UT
from LOS HS
Attends TAMU
from CS HS
0.603*
(0.344)
0.514*
(0.284)
-0.098
(0.137)
-0.020
(0.193)
0.067
(0.111)
0.447*
(0.234)
-0.018
(0.167)
0.333
(0.185)
-0.006
(0.014)
0.030
(0.022)
0.014
(0.061)
0.251**
(0.104)
Panel B: Upper Bound Treatment-on-Treated (2SLS) Estimates of CS/LOS Programs
Sample Mean
CS
LOS
Social
Science
(3)
-0.563*
(0.291)
-1.556***
(0.356)
0.559
-0.027
(0.020)
-0.038**
(0.019)
0.436
-0.031*
(0.017)
-0.076***
(0.017)
Other
Notes: Estimation of equations (1) and (4) in the text using the linked ERC-THECB data for the 1996-2002 high school
graduating cohorts. Each group of two coefficient estimates in each column comes from the same regression. All models
include high school and year fixed effects as well as the demographic, high school and test score controls discussed in
Section 4 of the text. Sample sizes for the top 30% college attendee sample is 28,665. First-stage estimates are shown
in Table ??. Standard errors clustered at the high school level are in parentheses: ***, **, * indicate significance at
the 1%, 5% and 10% levels, respectively.
Top 30%
Bottom 70%
0.232
Sample Mean
CS
0.035*
(0.020)
0.022*
(0.013)
LOS
Top 30%
Business
(2)
Panel A: Intention-to-Treat Estimates of CS/LOS Programs
Liberal
Arts
(1)
Treatment
HS Class
Achievement
Table 7: The Effect of Longhorn Opportunity and Century Scholar Programs on Last
Major Recorded - College Attendees
51
LOS
Top 30%
202,956
56,844
-0.019
(0.015)
0.017
(0.017)
26,711
0.039**
(0.019)
0.020
(0.018)
202,956
-0.005
(0.004)
0.004
(0.004)
61,800
-0.005
(0.008)
0.007
(0.008)
126,168
0.001
(0.010)
-0.003
(0.011)
55,080
0.011
(0.014)
0.007
(0.014)
Panel B: High School Graduates
89,654
-0.000
(0.005)
0.009
(0.006)
28,665
-0.005
(0.008)
0.007
(0.008)
Panel A: College Attendees
ITT Estimates
In 10+ Years
6+ Years
Earnings Sample
After HS Grad
113,969
0.001
(0.011)
-0.013
(0.013)
49,670
0.012
(0.014)
-0.008
(0.014)
51,852
-0.001
(0.015)
0.004
(0.020)
24,320
0.044**
(0.019)
0.006
(0.020)
10+ Years
After HS Grad
55,080
0.350
(0.425)
0.357
(0.540)
26,711
0.656**
(0.307)
0.495
(0.335)
49,670
0.356
(0.437)
-0.255
(0.557)
24,320
0.746**
(0.323)
0.296
(0.364)
Upper Bound TOT Estimates
6+ Years
10+ Years
After HS Grad
After HS Grad
Notes: Estimation of equations (1) and (4) in the text using the linked ERC-THECB data for the 1996-2002 high school graduating cohorts.
Each group of two coefficient estimates in each column comes from the same regression. All models include high school and year fixed
effects as well as the demographic, high school and test score controls discussed in Section 4 of the text. Earnings are adjusted for college
graduating cohort year and quarter fixed effects as discussed in the text. First-stage estimates are shown in Table ??. Standard errors
clustered at the high school level are in parentheses: ***, **, * indicate significance at the 1%, 5% and 10% levels, respectively.
Obs
-0.003
(0.003)
-0.004
(0.003)
Bottom 70%
CS
61,800
Obs
LOS
CS
-0.002
(0.005)
-0.004
(0.005)
Top 30%
LOS
89,654
Obs
CS
-0.002
(0.003)
-0.001
(0.004)
Bottom 70%
LOS
28,665
-0.005
(0.006)
-0.002
(0.006)
In 6+ Years
Earnings Sample
Obs
CS
Treatment
HS Class
Achievement
Table 8: The Effect of Longhorn Opportunity and Century Scholar Programs on Ln(Earnings)
52
CS
LOS
12,708
0.006
(0.031)
-0.005
(0.039)
-
-
Any College
(1)
12,708
0.019***
(0.006)
0.010
(0.012)
5,638
0.023*
(0.013)
0.011
(0.026)
UT
(2)
12,708
0.000
(0.009)
0.020*
(0.011)
5,638
-0.018
(0.015)
0.039*
(0.022)
Other 4-yr
(5)
2-yr
(6)
5,638
0.026
(0.025)
0.058**
(0.025)
5,638
-0.025
(0.035)
-0.130***
(0.030)
12,708
0.002
(0.011)
0.011
(0.016)
12,708
0.033
(0.021)
0.087****
(0.018)
12,708
-0.066
(0.041)
-0.123***
(0.040)
Panel B: High School Graduates
5,638
-0.006
(0.023)
0.025
(0.025)
Panel A: College Attendees
Enroll in
TAMU
ERU
(3)
(4)
12,708
-0.004
(0.010)
0.008
(0.012)
5,638
-0.006
(0.014)
-0.012
(0.025)
4-yr BA
Attainment
(7)
12,708
0.014
(0.014)
0.030*
(0.016)
5,638
0.044*
(0.023)
0.021
(0.023)
6-yr BA
Attainment
(8)
11,652
-0.037
(0.033)
0.041
(0.037)
5,361
0.038
(0.037)
0.056
(0.045)
6+ Years Post
HS Ln(Earnings)
(9)
10,621
-0.042
(0.033)
0.056
(0.036)
4,945
0.052
(0.043)
0.101**
(0.043)
10+ Years Post
HS Ln(Earnings)
(10)
Notes: Estimation of equations (1) and (4) in the text using the linked ERC-THECB data for the 1996-2002 high school graduating cohorts. “ERU” denotes “Emerging
Research University.” Each group of two coefficient estimates in each column comes from the same regression. All models include high school and year fixed effects as well
as the demographic, high school and test score controls discussed in Section 4 of the text. Standard errors clustered at the high school level are in parentheses: ***, **, *
indicate significance at the 1%, 5% and 10% levels, respectively.
Obs
Top 30%
Obs
LOS
Top 30%
CS
Treatment
HS Class
Achievement
Table 9: ITT Estimates Excluding Comparison Schools in Same County as Treated Schools Among Top 30% Students
53
CS
LOS
15,769
0.036***
(0.009)
-0.000
(0.011)
12,896
0.017**
(0.007)
-0.003
(0.008)
UT
(1)
15,769
-0.000
(0.006)
0.011
(0.007)
12,896
-0.011
(0.007)
0.024**
(0.011)
15,769
-0.022*
(0.012)
0.003
(0.014)
12,896
0.003
(0.012)
-0.012
(0.017)
Enroll in
TAMU
ERU
(2)
(3)
2-yr
(5)
12,896
-0.044*
(0.024)
-0.005
(0.021)
15,769
0.016
(0.020)
-0.013
(0.021)
15,769
-0.030
(0.025)
0.002
(0.020)
Panel B: Women
12,896
0.032
(0.021)
-0.005
(0.017)
Panel A: Men
Other 4-yr
(4)
15,769
-0.001
(0.010)
-0.032**
(0.013)
12,896
0.001
(0.010)
-0.032**
(0.011)
4-yr BA
Attainment
(6)
15,769
0.032*
(0.017)
-0.006
(0.016)
12,896
0.041**
(0.015)
-0.034**
(0.018)
6-yr BA
Attainment
(7)
14,844
0.013
(0.025)
0.037
(0.036)
11,867
0.078**
(0.028)
-0.005
(0.034)
6+ Years Post
HS Ln(Earnings)
(8)
13,606
0.022
(0.027)
0.022
(0.032)
10,714
0.074**
(0.032)
-0.015
(0.038)
10+ Years Post
HS Ln(Earnings)
(9)
Notes: Estimation of equation (1) in the text using the linked ERC-THECB data for the 1996-2002 high school graduating cohorts. “ERU” denotes
“Emerging Research University.” Each group of two coefficient estimates in each column comes from the same regression. All models include high school
and year fixed effects as well as the demographic, high school and test score controls discussed in Section 4 of the text. Standard errors clustered at the
high school level are in parentheses: ***, **, * indicate significance at the 1%, 5% and 10% levels, respectively.
Obs
Top 30%
Obs
LOS
Top 30%
CS
Treatment
HS Class
Achievement
Table 10: ITT Estimates by Gender Among Top 30 College Attendees